Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
JavaScript 3d-generationai-agentsanimationavatardigital-humangenerative-aiglbgltfimage-to-3dllmmcpmcp-servermodel-context-protocolsolanatext-to-3dthreejsweb-componentswebglwebxrx402
chore(prompts): drop the 102 finish prompts already executed (001-014, 105-199)
nirholasd33ad4ea319b
diff --git a/prompts/finish/001-swarm-100-sweep-authed-audit.md b/prompts/finish/001-swarm-100-sweep-authed-audit.mddeleted file mode 100644index 8cb848bf2..000000000--- a/prompts/finish/001-swarm-100-sweep-authed-audit.md+++ /dev/null@@ -1,86 +0,0 @@-# Sweep: Authenticated page sweep restored and clean--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--Run the sweep's own command first and work from its output, not from any-memory of what used to fail:-- npm run audit:web:login--Capture the full output to the session scratchpad; do not work from an-excerpt.--## Task--The QA credentials (AUDIT_EMAIL / AUDIT_PASSWORD) are absent from .env, Secret Manager, and the Cloud Run env, so the authed sweep has no working login. First provision a fresh real QA account through the production /register flow, store the credentials in .env and on the Cloud Run service env (update single vars with --update-env-vars, never --set-env-vars), then run the authed sweep and fix every failing page at root.--Fix everything found in this sweep's scope. Findings that belong to a-different surface entirely (and would take this session off its one task) go-in the report as named follow-ups instead; everything in scope gets fixed-here, at root, with no masking.--## Definition of done--- [ ] The command above exits 0 (or, for measurement-style sweeps, the report- carries the measured numbers and every committed fix).-- [ ] Every fix is a root-cause fix; no check was weakened, skipped, or- quarantined to get to green.-- [ ] npm run check:rules -- --paths <touched files> exits 0; npm test passes- if code changed.--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/001-swarm-100-sweep-authed-audit.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/002-quality-bar-07-design-system-sweep.md b/prompts/finish/002-quality-bar-07-design-system-sweep.mddeleted file mode 100644index dc110abf3..000000000--- a/prompts/finish/002-quality-bar-07-design-system-sweep.md+++ /dev/null@@ -1,101 +0,0 @@-# QB-07: Design-system consistency sweep (tokens, states, microinteractions)--**How to run this:** paste this whole file into a fresh Claude Code chat opened in-`/workspaces/three.ws`, or say "execute `prompts/finish/002-quality-bar-07-design-system-sweep.md`".-It is complete on its own. Also read `prompts/finish/_context/quality-bar-_shared.md`, `DESIGN-TOKENS.md`-and `CLAUDE.md`.--## Binding operating clause--1. Finish 100%. Never end with a question or an unexecuted plan. This is a sweep, so if the- session cannot cover every surface, finish complete surfaces cleanly and list the remainder- ordered by traffic. Never leave a surface half-migrated.-2. Blockers have pre-answered routes at the bottom.-3. CLAUDE.md hard rules: no mocks, no TODO comments, no commented-out code, no em-dash or- en-dash characters. Stage explicit paths only, commit per surface.--## Mission--Make 300+ pages feel like one product: one token sheet, consistent interactive states,-designed empty, error and loading states everywhere, and the microinteractions that signal-quality.--## Step 0: re-derive current state (trust nothing below)--```bash-npm run audit:tokens # scripts/audit-token-drift.mjs, the drift gate-npm run audit:console # console errors across pages-npm run audit:overlays-npm run audit:a11y # playwright axe pass on the top pages-grep -rn "#[0-9a-fA-F]\{6\}" src/ pages/ --include=*.css --include=*.html | wc -l-```--Record those numbers. They are your before-baseline and the report must show the after-numbers-next to them.--## Tasks--1. **Token audit.** Find the canonical token sheet (`src/styles/`, `DESIGN-TOKENS.md`).- Inventory hardcoded values across `src/`, `pages/`, `public/*.html` (colors, px spacing,- font sizes, radii, shadows, z-indices). Consolidate to tokens; where two near-identical- values exist, pick one canonical value and use it everywhere.-2. **Interactive-state pass.** Every button, link, card and input gets hover, active,- focus-visible and disabled states from shared classes and tokens. Focus rings must be- visible on dark and light contexts. Grep for `cursor: pointer` elements with no state rules- as a cheap detector.-3. **State design sweep.** For each major surface (forge, markets, marketplace, agents,- dashboard, wallet, launches, changelog, news, play, walk, irl, ar, scene studio, docs):- verify loading (skeletons matching the final layout, not spinners), empty (says what to do- with a working call to action), error (plain language plus a recovery action), and overflow- (1,000 items, 200-character names, tiny screens). Fix inline; keep a per-surface checklist.-4. **Motion language.** One standard: 150 ms micro, 250 ms panel, one easing set, opacity and- transform only, `prefers-reduced-motion` honored globally.-5. **Typography and rhythm.** One type scale and line-height rhythm; semantic heading- hierarchy (one h1 per page, no skipped levels); readable measure on text-heavy pages.-6. **Contrast and accessibility floor.** Fix everything `npm run audit:a11y` reports on the top- 30 pages: contrast below AA, missing labels, keyboard traps.--## Definition of done--- [ ] Token sheet canonical; a raw-hex grep over `src/` returns only the token sheet plus- justified exceptions, each listed in the report.-- [ ] Per-surface state checklist complete, with the fixes shipped.-- [ ] `npm run audit:tokens`, `npm run audit:console`, `npm run audit:a11y` all better than the- Step 0 baseline and none worse. Before and after numbers in the report.-- [ ] `npm run audit:web` shows no new errors; visual spot-check at 320, 768, 1440 px on the- top 10 pages.-- [ ] One `data/changelog.json` entry summarizing the visible polish in holder language.-- [ ] `npm run check:rules -- --paths <files you touched>` clean.--## Never blocked (pre-answered)--| Blocker | Do this |-|---|---|-| Scale: too many surfaces for one session | Batch by surface, commit per surface with explicit paths, and list the remainder ordered by traffic. Never leave one half-migrated. |-| A legacy page under `public/*.html` is not in the Vite graph | Remember the raw-`/src` trap: an unregistered page shipping `/src` imports breaks (a CSS import kills the page). Test every touched public page actually renders. |-| Light vs dark | The platform is dark-first. If a light context exists (docs, embeds), tokens carry both or explicitly declare dark-only. Do not invent a light theme here. |-| A token change touches a file another agent is editing | Re-read the file immediately before each edit and stage only your own paths. |--## Report format--Before and after audit numbers, the per-surface checklist, the exceptions list, and the-remainder ordered by traffic. No recap of this file.--## Retire this prompt when it is done (required)--1. Verify every Definition of done line against actual command output in front- of you. Never claim a line you did not verify.-2. Record the outcome in this campaign's PROGRESS or INDEX file if it has one.-3. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares), and delete this- prompt file in that same commit:-- git rm prompts/finish/002-quality-bar-07-design-system-sweep.md-- A finished order left on disk reads as open work to the next agent, so the- shrinking directory is the campaign's progress ledger.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns it.-Never delete this file on a partial.diff --git a/prompts/finish/003-quality-bar-04-pbr-texture-material-realism.md b/prompts/finish/003-quality-bar-04-pbr-texture-material-realism.mddeleted file mode 100644index 4fadeabd1..000000000--- a/prompts/finish/003-quality-bar-04-pbr-texture-material-realism.md+++ /dev/null@@ -1,107 +0,0 @@-# QB-04: PBR texture and material realism--**How to run this:** paste this whole file into a fresh Claude Code chat opened in-`/workspaces/three.ws`, or say "execute `prompts/finish/003-quality-bar-04-pbr-texture-material-realism.md`".-It is complete on its own. Also read `prompts/finish/_context/quality-bar-_shared.md` and `CLAUDE.md`.--## Binding operating clause--1. Finish 100%. Never end with a question, a plan you did not execute, or "should I proceed?".-2. Every blocker has a pre-answered route at the bottom. Use it and keep going.-3. CLAUDE.md hard rules: no mocks, no stubs, no TODO comments, no em-dash or en-dash- characters. Stage explicit paths only. Do not push unless asked.-4. GPU and Vertex spend is pre-approved. Never downgrade quality to save credits.--## Mission--A perfect mesh still looks fake with a flat albedo texture. Every GLB the platform produces-should carry a complete PBR set (albedo, normal, roughness/metallic, AO where bakeable) tuned-per material class, so skin reads as skin and metal reads as metal under the viewers'-image-based lighting.--## Step 0: re-derive current state (trust nothing below)--```bash-ls workers/texture workers/remesh workers/stylize-grep -rn "restyle_material" api/_mcp3d/ api/_mcp/ mcp-server/src/tools/ | head-gcloud run services list --region us-central1 --project aerial-vehicle-466722-p5 \- --format="value(metadata.name)" | grep -E "texture|remesh|stylize"-sed -n '1,40p' scripts/compress-glbs.mjs-```--Then generate one real GLB per live lane through `POST https://three.ws/api/forge` and inspect-which PBR channels each actually emits. `@gltf-transform/core` and `three` are already in the-dependency tree; do not add a library for this.--## Tasks--1. **Channel matrix.** Build `scripts/inspect-glb-materials.mjs` (if no equivalent exists) that- prints, per material: which of baseColor, normal, metallicRoughness, occlusion and emissive- textures exist, their resolutions, and the extension list. Run it over one real output per- lane. The matrix goes in the report and in `workers/texture/README.md`.-2. **Derive the missing maps.** Where a lane emits albedo only, add a post-stage in- `workers/texture/` (GPU, credits approved): normal-from-height estimation plus a- roughness/metallic inference pass keyed by the director's material classification- (`api/_lib/forge-director-prompts.js`). Wire it into the forge pipeline as a final, opt-out- stage, never a separate user step.-3. **Measured-value presets.** Extend `restyle_material`'s preset library with real-world- values: skin roughness 0.45 to 0.6, nonmetal metallic 0.0, brushed-steel anisotropy hints,- car paint via `KHR_materials_clearcoat`, glass via `KHR_materials_transmission`. Render each- preset in the platform viewer under every HDRI the viewer ships before shipping it.-4. **Skin, eyes, hair for avatars.** For `text_to_avatar` and `forge_avatar` outputs: subsurface- approximation (`MeshPhysicalMaterial` thickness/attenuation or a baked SSS tint), a separate- eye material with high specular and a clear cornea, hair with an alpha-cutout double-sided- setup. This is the difference between a person and a mannequin.-5. **Texture resolution by tier.** Free tier keeps current sizes; standard and high go to 2K/4K- albedo with matching normals. Keep payloads sane by extending the existing meshopt plus WebP- chain in `scripts/compress-glbs.mjs` rather than writing a second encoder. Verify a- 4K-textured GLB still loads under Playwright WebKit and Android Chrome emulation; if it does- not, tier the delivery (viewer fetches compressed, download offers full-res).-6. **Prove it.** Run the fixed benchmark set (`data/quality-bench/prompts.json`) through- `node scripts/quality-bench.mjs` before and after, covering skin, metal, glass and wood- subjects. `node scripts/quality-bench.mjs --compare=latest,previous` exits nonzero on a mean- drop greater than 1.0; that gate must pass.--## Definition of done--- [ ] Channel matrix documented per lane; every lane emits or gains a full PBR set.-- [ ] Presets verified in-viewer under all shipped HDRIs, with screenshots.-- [ ] Avatar skin/eye/hair path shipped and screenshotted at 320, 768, 1440 px.-- [ ] Mobile payload check done, with real byte sizes.-- [ ] Quality-bench comparison run and no regression; numbers in the report.-- [ ] `npm test` green; `data/changelog.json` entry; `workers/texture/README.md` updated.-- [ ] `npm run check:rules -- --paths <files you touched>` clean.--## Never blocked (pre-answered)--| Blocker | Do this |-|---|---|-| KTX2 or WebP encode tooling | Use `@gltf-transform` plus the encoder already wired in `scripts/compress-glbs.mjs`. Never hand-roll an encoder. |-| A new model download looks necessary | Prefer what the texture worker already ships. If unavoidable, pull it from `gs://three-ws-model-weights`, never a runtime hub fetch. |-| A preset looks wrong under one HDRI | Test under every environment preset the viewer ships before tuning; record which HDRI drove the value. |-| Vertex refuses a prompt | The reason is in `promptFeedback.blockReason`, not `finishReason`. Rephrase and continue. |-| A GPU worker is cold or scaling | Expected. Wait or raise minScale within quota; never switch to a weaker lane to save time. |--## Report format--The channel matrix, before and after renders, the quality-bench delta, real byte sizes, and any-single remaining owner action. No recap of this file.--## Retire this prompt when it is done (required)--1. Verify every Definition of done line against actual command output in front- of you. Never claim a line you did not verify.-2. Record the outcome in this campaign's PROGRESS or INDEX file if it has one.-3. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares), and delete this- prompt file in that same commit:-- git rm prompts/finish/003-quality-bar-04-pbr-texture-material-realism.md-- A finished order left on disk reads as open work to the next agent, so the- shrinking directory is the campaign's progress ledger.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns it.-Never delete this file on a partial.diff --git a/prompts/finish/005-quality-bar-08-mobile-performance.md b/prompts/finish/005-quality-bar-08-mobile-performance.mddeleted file mode 100644index f78079030..000000000--- a/prompts/finish/005-quality-bar-08-mobile-performance.md+++ /dev/null@@ -1,108 +0,0 @@-# QB-08: Mobile experience and asset performance--**How to run this:** paste this whole file into a fresh Claude Code chat opened in-`/workspaces/three.ws`, or say "execute `prompts/finish/005-quality-bar-08-mobile-performance.md`".-It is complete on its own. Also read `prompts/finish/_context/quality-bar-_shared.md` and `CLAUDE.md`.--## Binding operating clause--1. Finish 100%. Never end with a question or an unexecuted plan.-2. Blockers have pre-answered routes at the bottom.-3. CLAUDE.md hard rules: no mocks, no TODO comments, no em-dash or en-dash characters. Stage- explicit paths only. Publish only real measured numbers, never estimates.--## Mission--A crypto-native audience lives on phones. Make the platform genuinely excellent on a mid-tier-phone: fast first paint, smooth 3D, no WebGL exhaustion, small GLB payloads, thumb-friendly-controls.--## Step 0: re-derive current state (trust nothing below)--A measured baseline already exists. Read it, then re-measure, because production has moved:--```bash-sed -n '1,40p' prompts/quality-bar/_generated/08/baseline.md-sed -n '1,40p' prompts/quality-bar/_generated/08/touch-baseline.md-node scripts/mobile-perf.mjs # Playwright field-style metrics, NOT Lighthouse-npm run audit:mobile-touch # scripts/mobile-touch-audit.mjs-```--The 2026-07-29 baseline measured `/play` LCP 37.7 s, `/marketplace` 30.4 s, `/` 27.4 s,-`/forge` TBT 56 s on a throttled Pixel 5. Those are the pages that matter most; if your fresh-run disagrees, trust your run and say so.--Lighthouse is not installed and must not be added. `scripts/mobile-perf.mjs` measures LCP, CLS,-FCP and a long-task blocking-time proxy directly through `PerformanceObserver` under emulated-mobile hardware and network. Quote its numbers with that caveat attached, exactly as-`baseline.md` does.--## Tasks--1. **Fix the worst pages first, in baseline order.** For each of the top five worst LCP or TBT- pages: find the actual cause (main-thread blocking module, unlazy 3D, oversized image,- render-blocking fetch), fix it, re-measure the same page with the same harness, and record- the delta. A fix that does not move the number is not a fix.-2. **GLB delivery.** `scripts/compress-glbs.mjs` already runs dedup, prune, resample, quantize,- `EXT_meshopt_compression` and WebP textures. Make it the default on the serve path: models- written through `api/_lib/r2.js` callers get compressed at upload or finalize time, with the- full-res original still available on the download action. Target: a typical forge output- under 8 MB over the wire with no visible quality loss. Feature-detect decoder support and- fall back to uncompressed.-3. **WebGL context discipline.** No page may hold more than two live WebGL contexts. Verify on- real mobile viewports; grid pages use static thumbnails that upgrade to live 3D on tap. The- `GL made/live/visible` column in `mobile-perf` output is your evidence.-4. **Input ergonomics.** 44 px minimum touch targets, viewer gestures not fighting page scroll- (correct `touch-action`), primary actions reachable at the bottom of generation flows,- safe-area insets respected. `npm run audit:mobile-touch` must come back clean.-5. **Network resilience.** Generations survive tab backgrounding and flaky networks: poll with- backoff and resume, never lose a finished result because a socket dropped. Test by- throttling to slow 3G mid-generation and by backgrounding the tab.-6. **Re-measure everything and publish honestly.** Same harness, same pages, before and after- table. Only real numbers go in the changelog.--## Definition of done--- [ ] Before and after table for every top-15 page, produced by the same harness, in the report- and committed under `prompts/quality-bar/_generated/08/`.-- [ ] Compressed GLB path proven under Playwright WebKit and Android Chrome emulation, with the- real-device caveat stated.-- [ ] No page exceeds two live WebGL contexts; grid pages tap-to-activate.-- [ ] `npm run audit:mobile-touch` clean; `npm test` green.-- [ ] `data/changelog.json` entry quoting only measured numbers; docs updated where embed or- SDK behavior changed.-- [ ] `npm run check:rules -- --paths <files you touched>` clean.--## Never blocked (pre-answered)--| Blocker | Do this |-|---|---|-| Lighthouse is missing | Deliberate. Use `scripts/mobile-perf.mjs` and keep its caveat wording. Never add Lighthouse as a dependency. |-| `@gltf-transform` version conflicts with the tree's three.js | Pin the CLI as a devDependency and run it from scripts at build time; leave the runtime untouched. |-| KTX2 or WebP decode fails on an old device | Serve the transcoder assets locally (never hotlink a CDN) and fall back to png/jpg textures on decode failure. |-| Old GLBs already on R2 | Compress at write time for new assets and add a lazy migrate-on-first-serve for old ones. Never bulk-rewrite storage. |-| A page's slowness is caused by someone else's in-flight change | Re-measure after `git pull` is not an option here (shared worktree); measure against production instead and say which revision you measured. |--## Report format--The before and after table, the cause you found per page, the compression byte deltas, and any-single remaining owner action. No recap of this file.--## Retire this prompt when it is done (required)--1. Verify every Definition of done line against actual command output in front- of you. Never claim a line you did not verify.-2. Record the outcome in this campaign's PROGRESS or INDEX file if it has one.-3. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares), and delete this- prompt file in that same commit:-- git rm prompts/finish/005-quality-bar-08-mobile-performance.md-- A finished order left on disk reads as open work to the next agent, so the- shrinking directory is the campaign's progress ledger.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns it.-Never delete this file on a partial.diff --git a/prompts/finish/006-quality-bar-03-gpu-fleet-scaleout.md b/prompts/finish/006-quality-bar-03-gpu-fleet-scaleout.mddeleted file mode 100644index 4cf2cd039..000000000--- a/prompts/finish/006-quality-bar-03-gpu-fleet-scaleout.md+++ /dev/null@@ -1,154 +0,0 @@-# QB-03: GPU fleet scale-out and cold-start honesty--**How to run this:** paste this whole file into a fresh Claude Code chat opened in-`/workspaces/three.ws`, or say "execute `prompts/finish/006-quality-bar-03-gpu-fleet-scaleout.md`".-It is complete on its own. Also read `prompts/finish/_context/quality-bar-_shared.md` and `CLAUDE.md`.--## Binding operating clause--1. Finish 100%. Never end this session with a question, a plan you did not execute, or- "should I proceed?". The owner is not available and does not want to be asked.-2. Every blocker has a pre-answered route in "Never blocked" at the bottom. Use it, note what- you did in the report, keep going. If one part is genuinely impossible in this environment,- ship every other part in full and name the single missing action in the report.-3. CLAUDE.md hard rules apply: no mocks, no stubs, no TODO comments, no fake progress, no- em-dash or en-dash characters anywhere you write. Stage explicit paths only; other agents- share this worktree. Do not `git push` unless the owner's message asked for it.-4. GCP spend is pre-approved (credits). Never trade quality for cost.--## Session status as of 2026-09-08 (read before re-running anything)--A prior session executed this order. Do not redo tasks 2 and 3.--- **Task 2 (cold-start honesty): DONE and verified.** `/forge` renders the boot state at- 320/768/1440 with zero console errors, its countdown uses the lane's real- `cold_start_seconds`, and its elapsed meter is now anchored to the server's- `elapsed_seconds` instead of a tab-local clock. `/api/gpt-forge` now returns- `cold_start` on POLL frames (it only did so at submit, which is why the MCP- surfaces, who see nothing but poll frames, could never report a boot), and the- `_mcp-studio` tools narrate it. The wording lives once in- `src/shared/forge-frames.js` (`coldStartState` / `coldStartLabel`), used by the- homepage chamber, the Studio, the in-world prop forge and the MCP tools.-- **Task 3 (keep-warm): was ALREADY shipped before this order was written.**- `/api/cron/gpu-keepwarm` runs on exactly the `*/10 14-23,0-4` schedule this file- asks for, and is deliberately quota-aware rather than warming every minScale=0- worker (the us-central1 L4 grant is 3 and two floors already hold 2; pinning a- third starved `model-text2motion` on 2026-07-26). The gap that was fixed:- `model-hunyuan3d`, the fleet's longest spin-up, was missing from the registry- entirely, so no override could ever reach it. `tests/cron-forge-lane-guards.test.js`- now fails if a routed scale-to-zero lane is absent or carries the wrong url env.-- **Task 1 (scale ceilings): BLOCKED on one `gcloud auth login`.** Also note its target- numbers conflict with a twice-learned lesson recorded in `docs/ops/gcp-credits-plan.md`:- at a granted L4 quota of 3, `model-trellis` min 1 + `model-hunyuan3d` min 1 + `model-rig`- min 1 consume the whole pool and trellis can never burst to its stated max 3. Read the- live grant before applying them; if it is still 3, raise the grant first.-- **Tasks 4 and 5 (load test, prove a lane): BLOCKED by a production outage, not by capacity.**- Every `/api/forge` generation fails with `SignatureDoesNotMatch` from object storage: the- R2 credential on the Cloud Run service is rejected, so no lane can park a reference image- or a finished mesh. 24h metrics read 12 attempts, 0 successful. Only a pre-outage CACHED- result still returns. There is nothing to load test until the credential is rotated and the- pending commits deploy (the graceful-degradation fix for this exact class landed- 2026-09-07 and is not yet in production, which is why the raw AWS signing sentence is- currently shown to users).--## Step 0: re-derive current state (do this first, trust nothing below)--This file's claims rot. Measure before you plan:--```bash-gcloud run services list --region us-central1 --project aerial-vehicle-466722-p5 \- --format="table(metadata.name, status.conditions[0].status)"-timeout 120 gcloud alpha quotas preferences list --project=aerial-vehicle-466722-p5 2>&1 | head -40-gcloud run services describe three-ws-api --region us-central1 \- --project aerial-vehicle-466722-p5 --format=yaml \- | grep -E "GCP_(TRIPOSG|TEXT2MOTION|HUNYUAN3D)_URL|MODEL_TRELLIS_URL"-curl -s "https://three.ws/api/forge?health" | head -40-```--As of 2026-08-01, `model-triposg`, `model-text2motion`, `model-trellis`, `model-triposr`,-`model-hunyuan3d` and `model-rig` were all Ready, so the original work order's "fix TripoSG"-and "deploy text2motion" tasks are shipped. Confirm that yourself, record the evidence, and-spend the session on what is still open.--The quota listing is slow and has returned empty in this workspace. Do not let it stall you:-`docs/ops/gcp-credits-plan.md` records the fleet position and the pre-approved scaling, and an-update that exceeds the grant fails with an explicit quota error, which is itself a measurement.-Attempt the scale change, read the error if there is one, and report the real ceiling.--## Tasks--1. **Scale ceilings.** Read the granted L4 quota- (`NvidiaL4GpuAllocNoZonalRedundancyPerProjectRegion`). Within the grant, set- `model-trellis` min 1 / max 3, `model-hunyuan3d` min 1 / max 2, `model-rig` max 2,- `model-triposr` min 0 / max 2. Verify each service stays Ready after the update- (`gcloud run services describe <name> --region us-central1 --format="value(status.conditions[0].status)"`).- If the grant is smaller than the plan needs, scale to what fits, file a fresh preference- at a smaller ask, and record both.-2. **Cold-start honesty in every polling surface.** `api/_lib/forge-lane-health.js` already- computes a cold-start ETA. Every surface that polls a generation must render a real state:- which lane, "GPU waking up, about Ns" with elapsed and remaining, then live stage labels- (queued, shape, texture, finishing). Surfaces to check and fix: `/forge`- (`pages/forge.html`, `src/forge*.js`), `/create`, MCP tool responses in `api/_mcp3d/` and- `api/_mcp-studio/`, and any agent screen showing generation progress. Skeletons, never- spinners. No fake progress: every number comes from the API response.-3. **Keep-warm only where minScale is 0.** For each such worker add a Cloud Scheduler healthz- ping during peak hours (14:00 to 04:00 UTC, `*/10`). Crons are Cloud Scheduler jobs synced- from `vercel.json`'s `crons` array by `scripts/create-gcp-scheduler.mjs`; follow an existing- job's shape and `docs/ops/gcp-production.md`. Never propose GitHub Actions.-4. **Load test through the router, not the workers.** Fire 10 concurrent real generations at- `POST https://three.ws/api/forge` across mixed tiers, poll each to a finished GLB, and- record queue behavior, failover events, p50 and p95 completion. A job that dies silently is- a defect: fix it (`api/_lib/forge-scale.js` owns concurrency limits) and re-run.-5. **Prove a lane, do not assume it.** A lane counts as live only when a real prompt returned a- real GLB through `/api/forge` and that GLB parses with geometry. Direct worker curl does not- count. Save the returned URLs in the report.--## Definition of done--- [ ] Scale ceilings applied within quota (or a smaller-ask request filed and recorded); every- touched service still Ready.-- [ ] Cold-start and stage states render real data on `/forge` at 320, 768 and 1440 px,- verified in a real browser via `npm run dev`, zero console errors from your code.-- [ ] Keep-warm scheduler jobs exist for every minScale=0 GPU worker, listed in the report.-- [ ] Load-test table (lane, tier, p50, p95, failovers, dropped jobs) with zero silent drops- after your fixes.-- [ ] `npm test` green (never pipe it through `tail`, that masks the exit code).-- [ ] `data/changelog.json` entry for the visible speed and reliability gain, holder language,- validated by `npm run build:pages`.-- [ ] `npm run check:rules -- --paths <files you touched>` clean.--## Never blocked (pre-answered)--| Blocker | Do this |-|---|---|-| `gcloud` auth dead (`invalid_rapt`) | It is the sperax.io Workspace reauth policy, not token expiry. Do every non-gcloud task in full, put the exact gcloud commands in the report as ready-to-run one-liners, and note that one `gcloud auth login` clears them. |-| Quota stuck reconciling | Proceed with everything else; the scale-up commands go in the report. |-| Worker weights missing | Stage from `gs://three-ws-model-weights` or the worker README's source. Staging weights is part of the task. |-| A worker will not bind in time | Use the bind-first plus background-load pattern from `workers/model-trellis/main.py`, plus `--timeout=600` and a generous startup probe. |-| Scheduler job needs a service account | Use the `three-ws@` runtime SA pattern from `docs/ops/gcp-production.md`. |-| A third-party lane is down | Every lane has a failover chain in `api/forge.js`. Use it; adding a missing rung is part of the task. |--## Report format--What shipped, what was already shipped before you started plus the evidence that proved it,-the load-test table, every command that changed cloud state, and the single owner action if one-remains. No recap of this file.--## Retire this prompt when it is done (required)--1. Verify every Definition of done line against actual command output in front- of you. Never claim a line you did not verify.-2. Record the outcome in this campaign's PROGRESS or INDEX file if it has one.-3. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares), and delete this- prompt file in that same commit:-- git rm prompts/finish/006-quality-bar-03-gpu-fleet-scaleout.md-- A finished order left on disk reads as open work to the next agent, so the- shrinking directory is the campaign's progress ledger.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns it.-Never delete this file on a partial.diff --git a/prompts/finish/007-quality-bar-10-avatar-likeness-irl-people.md b/prompts/finish/007-quality-bar-10-avatar-likeness-irl-people.mddeleted file mode 100644index e7b04c117..000000000--- a/prompts/finish/007-quality-bar-10-avatar-likeness-irl-people.md+++ /dev/null@@ -1,111 +0,0 @@-# QB-10: Avatar likeness, IRL people quality--**How to run this:** paste this whole file into a fresh Claude Code chat opened in-`/workspaces/three.ws`, or say "execute `prompts/finish/007-quality-bar-10-avatar-likeness-irl-people.md`".-It is complete on its own. Also read `prompts/finish/_context/quality-bar-_shared.md` and `CLAUDE.md`.--## Binding operating clause--1. Finish 100%. Never end with a question or an unexecuted plan.-2. Blockers have pre-answered routes at the bottom.-3. CLAUDE.md hard rules: no mocks, no TODO comments, no em-dash or en-dash characters. Stage- explicit paths only. Never add a rig allowlist; fixes go at the mapping layer.--## Mission--Photo-to-avatar and text-to-avatar outputs should look like real people: correct proportions,-believable skin, hands with five fingers, clothes with fabric behavior, and a rig that animates-without breaking the illusion.--## Step 0: re-derive current state (trust nothing below)--```bash-node scripts/animation-dignity-sweep.mjs --verbose # ten bone-naming conventions, measured-ls prompts/quality-bar/_generated/10/ # committed evidence from the last pass-npx vitest run tests/glb-canonicalize.test.js tests/animation-retarget.test.js-gcloud run services list --region us-central1 --project aerial-vehicle-466722-p5 \- --format="value(metadata.name)" | grep -E "avatar-reconstruction|model-rig|hunyuan"-```--**Already proven, do not redo:** task 5 (animation dignity) is measured and green. Ten rig-conventions animate both arms and both legs down both production paths; the fixes landed in-`src/glb-canonicalize.js` with regression cover in both test files, and the evidence is-committed at `prompts/quality-bar/_generated/10/animation-dignity-sweep.{txt,json}`.--**The clavicle-pivot defect that pass left open is also fixed** (re-verified 2026-08-01): a-Rigify rig used to bind the `LeftArm`/`RightArm` clip track to `shoulder.L`/`shoulder.R`, the-clavicle, on the runtime-only lane. `resolveArmShoulderCollisions` now lives in-`src/glb-canonicalize.js:475` and both lanes call it (ingest through `canonicalizeJointNodes`,-runtime through `canonicalBoneEntries` in `src/animation-retarget.js:121`), so the sweep passes-with no warning. If a new rig ever trips it, widen that resolver; never add a per-rig name-special-case and never add a rig allowlist.--So the rig half of this work order is done. **What is open is the likeness half:** everything-below except the sweep.--## Tasks--1. **Audit the avatar path end to end** with 6 real cases: four text prompts (athlete,- grandmother, stylized kid, businessperson) and two CC0 reference photos. Score each against- the IRL bar: proportions, face fidelity, hands, feet, clothing, texture seams at the neck- and hairline. Document per-case failures precisely, with screenshots.-2. **Route avatars through the strongest live lane.** Confirm which mesh lane is Ready, prefer- the highest-quality one with portrait realism cues, and verify the humanoid gate still hands- its output to the rig worker cleanly with a normalized bind pose.-3. **Face fidelity for photo input.** Feed the actual user photo (multi-view when several were- given) to the image-to-3D lane rather than a regenerated lookalike, and keep the face- region's texture resolution highest. Never generate an avatar of a named third party from- text alone; photo input implies consent by upload, text-only real-person names keep the- generic treatment.-4. **Skin, eye and hair materials for avatars.** If QB-04 already shipped the material module,- wire it. Do not duplicate it.-5. **The `/irl` moment.** Place three finished avatars in AR through the animated USDZ bake- path and screenshot each. The pin should read as a person standing there, not a figurine.-6. **Publish.** Before and after gallery for all six cases through the whole chain (mesh,- materials, rig, animation, AR), `data/changelog.json` entry in holder language, and an- update to the `create-3d-avatar` skill guidance if defaults changed.--## Definition of done--- [ ] `node scripts/animation-dignity-sweep.mjs` still reports 10/10 on both lanes after your- changes (it does today; a regression here is a release blocker).-- [ ] 6 of 6 test cases visibly improved, gallery in the report, hands specifically called out- (count the fingers in the screenshots).-- [ ] AR screenshots included; humanoid-gate degrade path kill-tested (force a rig failure and- confirm the mesh still ships with a working fallback, never a T-pose).-- [ ] `npm test` green; `npm run audit:rig-coverage` clean.-- [ ] `data/changelog.json` entry; `npm run check:rules -- --paths <files you touched>` clean.--## Never blocked (pre-answered)--| Blocker | Do this |-|---|---|-| The strongest mesh lane is not Ready | Run the full audit and material, rig and AR work on the lane that is. The swap is one env var and a re-run; note it as the follow-up command. |-| Hands stay bad at max budgets | The honest mitigations are pose choice (relaxed hands), higher hand visibility in the reference views, and the director explicitly requesting "hands visible, fingers separated". Document what moved the needle with screenshots. |-| A stylized kid case drifts photoreal | Drop the case from public galleries and say so. Never ship photoreal likenesses of minors. |-| A new bone-naming convention appears | Add its mapping to `src/glb-canonicalize.js` plus a case in `tests/glb-canonicalize.test.js`. Never add a curated rig allowlist. |-| A GPU lane is cold | Wait it out. Do not fall back to a weaker lane for the quality audit. |--## Report format--The per-case scoring table with before and after images, the sweep result, the AR screenshots,-and any single remaining owner action. No recap of this file.--## Retire this prompt when it is done (required)--1. Verify every Definition of done line against actual command output in front- of you. Never claim a line you did not verify.-2. Record the outcome in this campaign's PROGRESS or INDEX file if it has one.-3. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares), and delete this- prompt file in that same commit:-- git rm prompts/finish/007-quality-bar-10-avatar-likeness-irl-people.md-- A finished order left on disk reads as open work to the next agent, so the- shrinking directory is the campaign's progress ledger.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns it.-Never delete this file on a partial.diff --git a/prompts/finish/008-swarm-100-sweep-i18n.md b/prompts/finish/008-swarm-100-sweep-i18n.mddeleted file mode 100644index 7b93f10dc..000000000--- a/prompts/finish/008-swarm-100-sweep-i18n.md+++ /dev/null@@ -1,86 +0,0 @@-# Sweep: i18n lint clean--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--Run the sweep's own command first and work from its output, not from any-memory of what used to fail:-- npm run i18n:lint--Capture the full output to the session scratchpad; do not work from an-excerpt.--## Task--Run the i18n lint and fix every finding: missing keys, unused keys, malformed interpolations. Where a hardcoded user-facing string should be a key, extract it with the existing i18n tooling.--Fix everything found in this sweep's scope. Findings that belong to a-different surface entirely (and would take this session off its one task) go-in the report as named follow-ups instead; everything in scope gets fixed-here, at root, with no masking.--## Definition of done--- [ ] The command above exits 0 (or, for measurement-style sweeps, the report- carries the measured numbers and every committed fix).-- [ ] Every fix is a root-cause fix; no check was weakened, skipped, or- quarantined to get to green.-- [ ] npm run check:rules -- --paths <touched files> exits 0; npm test passes- if code changed.--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/008-swarm-100-sweep-i18n.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/009-roadmap-generation-suite.md b/prompts/finish/009-roadmap-generation-suite.mddeleted file mode 100644index a4b26d0e2..000000000--- a/prompts/finish/009-roadmap-generation-suite.md+++ /dev/null@@ -1,128 +0,0 @@-# RM-GEN: Generation suite parity and production truth--**How to run this:** paste this whole file into a fresh Claude Code chat opened in-`/workspaces/three.ws`, or say "execute `prompts/finish/009-roadmap-generation-suite.md`".-It is complete on its own. Also read `prompts/finish/_context/roadmap-00-README.md` (safety doctrine and the-regression gate) and `CLAUDE.md`.--## Binding operating clause--1. Finish 100%. Never end with a question, a plan you did not execute, or "should I proceed?".-2. "Works" means verified in production by running it, after deploy. Tests passing is not the- bar; a stranger getting a GLB is the bar.-3. Fix the smallest broken thing that unblocks a whole flow before building anything new.-4. CLAUDE.md hard rules: no mocks, no stubs, no TODO comments, no em-dash or en-dash- characters. Stage explicit paths only. Deploys are owner-gated: prepare them so the ship is- one command.--## Mission--A generation suite that matches Meshy and Tripo on quality and beats them on agent-nativeness,-and that actually works in production. The repo has the surface area; the gap has historically-been between "code exists" and "a user can run it on three.ws today".--## Step 0: rebuild the truth table (mandatory, replaces every claim below)--The June 2026 audit that this file used to carry is stale: several workers that it listed as-undeployed are Ready now. Do not read old status from anywhere. Probe production yourself and-write the table into your report:--```bash-gcloud run services list --region us-central1 --project aerial-vehicle-466722-p5 \- --format="table(metadata.name,status.conditions[0].status)"-gcloud run services describe three-ws-api --region us-central1 \- --project aerial-vehicle-466722-p5 --format=yaml | grep -E "^ *- name: (GCP_|MODEL_|LONGCAT_|REPLICATE_)" -A1-curl -s "https://three.ws/api/forge?catalog" | python3 -m json.tool | head -60-curl -s "https://three.ws/api/forge?health" | python3 -m json.tool | head -60-```--Then, for each row, run the real flow end to end against production and record the outcome:--| Feature | Endpoint | Probe |-|---|---|---|-| Text to 3D | `POST /api/forge` | submit, poll to `done`, download the GLB, assert it parses with geometry |-| Image to 3D | `POST /api/forge` | presign, upload, submit, poll, download |-| Auto-rig | `POST /api/forge?action=rig` | rig a real mesh, assert bones and skin weights exist |-| Remesh / retopo | `POST /api/forge-remesh` | real mesh in, real mesh out |-| Stylize | `POST /api/forge-stylize` | same |-| Segment | `POST /api/forge-segment` | same |-| Background removal | `POST /api/forge-rembg` | real image in, alpha out |-| Text to motion | `POST /api/forge-motion` | clip out, plays on the default rig |-| Retexture | `POST /api/studio/retexture-*` | textures actually change |-| Talking avatar | the longcat worker route | real video out |-| x402 paid generation | `POST /api/x402/forge` | 402, pay, artifact |-| MCP 3D Studio tools | `POST /api/mcp-studio`, `/api/mcp-3d` | `tools/list` then a real `tools/call` |--A `configured: true` in the catalog means an env var exists, nothing more. That is exactly how-dead flows once looked green. Only a real artifact counts.--## Tasks--1. **Close every red row from Step 0.** Root-cause each (missing env var, worker not Ready,- provider regression, schema drift), fix it, and re-run the same probe. If the fix needs a- deploy, stage everything and name the one command.-2. **Health that tells the truth.** `GET /api/forge?health` must live-probe each provider- (cheap authenticated HEAD or equivalent) rather than reporting on env-var presence. Add or- extend the scheduled smoke test so a draft generation runs daily and alerts on failure.-3. **Surface the hidden tools in `/forge`.** Retexture (full and region), remesh, stylize,- segment, rig, animate and export exist as APIs or MCP tools; the result panel should offer- them as one pipeline on one page. Every button wired to the real endpoint.-4. **Preview then refine.** Cheap fast preview, then a one-click refine of the chosen result at- a higher tier conditioned on the same seed and reference. The tier system already exists.-5. **Export formats.** The remesh worker already converts glb, obj, stl, ply, usdz, 3mf and fbx.- Expose a format picker on download instead of GLB-only.-6. **PBR material controls.** Expose the map outputs (albedo, normal, roughness, metallic) and- a re-bake option in the result panel. Coordinate with `prompts/quality-bar/04-...` rather- than duplicating it.-7. **Job webhooks and public API docs.** Extend the existing Replicate webhook plumbing to- forge jobs and publish the endpoint contract in `docs/` with runnable examples, since we- sell it through x402.-8. **Community gallery.** Add an opt-in public showcase over the existing private gallery, with- a remix action that generates a variation.--## Definition of done--- [ ] Step 0 table filled in from real production probes, every row green or explained with a- named blocker and a one-command fix.-- [ ] `/api/forge?health` live-probes providers; the daily smoke test covers what you shipped.-- [ ] Every newly surfaced tool clicked through in a real browser at 320, 768 and 1440 px, with- loading, empty, error and success states designed.-- [ ] `npm run gate` no worse than the baseline you captured at the start; `npm test` green.-- [ ] `data/changelog.json` entries for the user-visible additions; docs updated for the API- contract.-- [ ] `npm run check:rules -- --paths <files you touched>` clean.--## Never blocked (pre-answered)--| Blocker | Do this |-|---|---|-| A worker is not deployed | Deploy it from `workers/<name>/cloudbuild.yaml` with the pinned build and runtime service accounts. Staging weights from `gs://three-ws-model-weights` is part of the task. |-| An env var is missing | Look in `.env`, then the Cloud Run service, then Secret Manager. Update single vars with `--update-env-vars`; `--set-env-vars` replaces the entire set. |-| A third-party provider 404s or throttles | Use the failover chain in `api/forge.js`; if a rung is missing, adding it is part of the task. |-| `gcloud` auth dead | Do everything else, then put the exact commands in the report; one `gcloud auth login` clears it. |-| A deploy is required | Owner-gated. Prepare it, name the single command, and finish everything else. |-| Production is behind `main` | Check `curl -s https://three.ws/api/version` against `git log` before debugging a "broken" feature. A stale revision explains most phantom failures. |--## Report format--The Step 0 truth table (before and after), the root cause per red row, what you shipped, and-the single owner action if one remains. No recap of this file.--## Retire this prompt when it is done (required)--1. Verify every Definition of done line against actual command output in front- of you. Never claim a line you did not verify.-2. Record the outcome in this campaign's PROGRESS or INDEX file if it has one.-3. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares), and delete this- prompt file in that same commit:-- git rm prompts/finish/009-roadmap-generation-suite.md-- A finished order left on disk reads as open work to the next agent, so the- shrinking directory is the campaign's progress ledger.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns it.-Never delete this file on a partial.diff --git a/prompts/finish/011-event-02-play-polish-sweep.md b/prompts/finish/011-event-02-play-polish-sweep.mddeleted file mode 100644index 87ab1a3f0..000000000--- a/prompts/finish/011-event-02-play-polish-sweep.md+++ /dev/null@@ -1,105 +0,0 @@-# 02 · /play polish sweep: the in-world half--**How to run:** paste this whole file into a fresh Claude Code chat in the three.ws repo, or tell the agent to read `prompts/finish/011-event-02-play-polish-sweep.md`. Read [00-CONTEXT.md](_context/event-00-CONTEXT.md) first.--**Operating clause:** finish 100%. Never end the session with a question, an unexecuted plan, or "let me know". CLAUDE.md hard rules bind: no mocks, no TODO comments, no em-dash characters, commits stage explicit paths only, pushes and production deploys are owner-gated. This surface is under heavy concurrent development: re-read a file immediately before each edit and stage only your own paths.--Rewritten 2026-09-02 to its remainder. The lobby half is verified and is not yours to redo. The in-world half is still owed, and the reason it is still owed is now a measured machine condition rather than a mystery. Read "The one precondition" before you run anything.--## What is already verified (do not redo)--- **The lobby, twice.** The 2026-08-08 preflight measured 130 desktop / 91 mobile tab stops with 0 focus rings missing and 0 controls refusing focus, 31 coin cards, 7 presets, create modal and gallery both closing on Escape, search resolving to hits and to a designed empty state, CLS 0.0065, no horizontal overflow at 1440 or 375. A 2026-09-02 re-run agreed: `[focus:lobby] 114 tabbable, 0 kinds without a focus ring, 0 that refuse focus`, 21 cards, 0 skeletons.-- **The world boots headless on this box.** This retires the residual the preflight and the order-04 agent both recorded ("the 3D scene will not boot headless under this load"). On 2026-09-02 `scripts/play-journey-audit.mjs` drove a real cold load, the lobby, search, the create modal, the gallery, world entry into the $THREE world, the HUD, and the store, bank, wheel and emote panel probes, in headless Chromium under `--use-gl=swiftshader --enable-unsafe-swiftshader`. The path works. There is no need to invent a new harness.--## The one precondition, and it is the whole reason this order is still open--**Run this on a quiet box.** `scripts/play-journey-audit.mjs` measures with wall-clock waits, so CPU contention does not slow it down, it makes it lie. Measured on 2026-09-02:--- At load average 43 on 16 cores the run reached the world but the shared Vite dev server on port 3000 died at +67s, and every subsequent line was `net::ERR_CONNECTION_REFUSED`. A store panel that "failed to load its module" and a `LOADER NEVER CLEARED` were both artifacts of a dead server, not defects.-- At load average 220 the same harness printed `LOBBY NEVER BECAME VISIBLE` at +125s and then `[state:lobby-first-paint] {"cards":21}` at +211s, and printed `GRID NEVER RESOLVED (no cards, no empty state, no error state)` about a grid holding 21 cards. Every verdict in that run is a false negative.--So, before you start and again when you finish:--```bash-uptime # 1-minute load must be well under 16-curl -s -o /dev/null -w '%{http_code}\n' http://localhost:3000/play # must be 200-curl -s -o /dev/null -w '%{http_code}\n' http://127.0.0.1:2567/health # world server, must be 200-```--If the load is high, wait it out or take the box quiet; do not run the sweep and do not report its output. If either server is down, `npm run dev:walk-all` boots both. A run whose tail shows a burst of `ERR_CONNECTION_REFUSED` on `localhost:3000` is a dead-server run: discard it and rerun, never triage its findings.--## Step 0 · Re-derive the current state--```bash-git log --oneline -15 -- src/game/-git status --short src/game/-```--`src/game/photo-mode.js` and its neighbours were being actively edited on 2026-09-02 under order 06. Anything dirty or freshly committed there belongs to that work.--## The run--```bash-TAB_CHECK=1 node scripts/play-journey-audit.mjs # desktop 1440x900-VIEWPORT=375 node scripts/play-journey-audit.mjs # phone-VIEWPORT=320 node scripts/play-journey-audit.mjs # narrowest supported-```--Each takes minutes even on a quiet box. Capture all three to files and build the defect list from them, then reproduce each finding by hand in a headed browser before you change any code: the harness locates defects, it does not diagnose them.--Two known harness readings to interpret rather than file:--- `[overflow:panel:*]` rows for `div.cc-label.ac-name`, `div.ac-prompt`, `span.ac-key`, `div.tik-prompt`, `div.npc-prompt` and `div.cc-label.npc-name` are billboard labels positioned in world space, so far-negative and far-positive `left` values are how they are supposed to behave when their subject is off camera. Confirm that reading once, then treat those selectors as noise. A DOM panel overflowing is a real defect; a world label at x=5746 is not.-- `[panel:bank] no trigger visible` is correct. The bank is proximity-gated behind its ATM by design ("the bank makes you walk there", changelog 2026-08-08), so it has no HUD button. Same for the Wheel of Fortune, which is gated behind its station; the harness excludes the emote wheel from that probe on purpose. Reaching those two panels means walking the avatar to the station, not clicking a HUD button.--## Tasks--Work the defect list worst first. The bar, with the usual owner of each class:--1. **Console hygiene in-world:** zero errors and zero warnings from our code across world entry, movement, chat, emotes, the store, the bank, the wheel, the jobs board and the friends drawer. Root-cause every one; never filter.-2. **Every in-world async surface has a designed loading, empty and error state.** Panel UIs live in [src/game/economy-ui.js](../../src/game/economy-ui.js), [src/game/quests-ui.js](../../src/game/quests-ui.js), [src/game/spin-wheel-ui.js](../../src/game/spin-wheel-ui.js), [src/game/friends-panel.js](../../src/game/friends-panel.js); the HUD chrome is in [src/game/coincommunities-ui.js](../../src/game/coincommunities-ui.js). Empty states say what to do next; error states say how to recover.-3. **Interactive-state audit in-world:** every panel control has hover, active and focus-visible states and is keyboard reachable, and Escape closes every panel that opens.-4. **Mobile at 375 and 320:** nothing overlaps, nothing overflows, touch targets 40px or larger, and the joystick, chat, emote and HUD stack still leave the world visible.-5. **Copy pass:** every string a player sees in-world reads like a product, not a log line.-6. For anything a player would notice, one collective `data/changelog.json` entry (tags: `improvement`), then `npm run build:pages`.--## Definition of done--- [ ] Three clean harness runs (desktop, 375, 320) captured on a quiet box, each with a live dev server at the start and the end of the run.-- [ ] The defect list is empty or every remaining item is in the report with a reason it was out of scope (in-flight elsewhere, server-side, needs owner).-- [ ] Every fix reproduced by hand in a headed browser before and after.-- [ ] `npm run check:rules -- --paths <files you touched>` passes; `npm run test:core` passes; any existing suite covering a file you touched passes.-- [ ] Changelog entry present; committed with explicit paths; [PROGRESS.md](_context/event-PROGRESS.md) appended.--## Never blocked--| Blocker | Resolution |-|---|---|-| Load average too high to measure | Wait, or run it later in the session. Reporting a contended run as a defect list is worse than reporting nothing; the 2026-09-02 attempt is the cautionary example. |-| Dev server dies mid-run | It is shared with every other agent on this worktree. Restart with `npm run dev:walk-all`, discard the poisoned run, rerun. |-| A defect lives in a file another agent has dirty in `git status` | Fix only if trivial and non-overlapping; otherwise record it as in-flight elsewhere and move on. |-| A fix needs a server change in `multiplayer/` | In scope. Room handlers are in `multiplayer/src/rooms/WalkRoom.js`; the module tests under `tests/` show the patterns. |-| Playwright missing browsers | `npx playwright install chromium`. |--## Report format--The three run files, the defect list with each item marked fixed / out-of-scope-and-why, the load average at the start and end of each run, test output verbatim, files committed.--## Retire this prompt when it is done (required)--1. Verify every Definition of done line against actual command output in front- of you. Never claim a line you did not verify.-2. Record the outcome in this campaign's PROGRESS or INDEX file if it has one.-3. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares), and delete this- prompt file in that same commit:-- git rm prompts/finish/011-event-02-play-polish-sweep.md-- A finished order left on disk reads as open work to the next agent, so the- shrinking directory is the campaign's progress ledger.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns it.-Never delete this file on a partial.diff --git a/prompts/finish/013-production-100-05-unwired-guards.md b/prompts/finish/013-production-100-05-unwired-guards.mddeleted file mode 100644index de064bf83..000000000--- a/prompts/finish/013-production-100-05-unwired-guards.md+++ /dev/null@@ -1,108 +0,0 @@-# P100-05: Guards that exist but run nowhere--**How to run this:** paste this whole file into a fresh Claude Code chat opened in-`/workspaces/three.ws`, or say "execute `prompts/finish/013-production-100-05-unwired-guards.md`".-Read [00-INDEX.md](_context/production-100-00-INDEX.md) and `CLAUDE.md` first.--## Why this order exists--On 2026-09-04 production answered `/api/healthz`, `/api/wk`, `/api/x402-pay` and every-`/api/home/*` route with a 500 for roughly thirteen hours. The cause was a packaging rule:-`.gcloudignore` is an allowlist and never re-included `services/`, so two modules the server-imports at load time were absent from the image. Nothing was red. A missing re-include is not-a build error, it is a runtime one.--`scripts/check-gcloudignore.mjs` had been written specifically to catch that, and its own-commit message claimed it "fails before a build instead of in production". It did not, because-it was referenced by no npm script. An unwired guard is not a guard, and the commit that-added it read as protection while providing none.--That guard is now wired into `gate` and `deploy:gcp:submit`, and-`tests/gcloudignore-wiring.test.js` pins the wiring. This order handles the rest of the class:-**39 of the repo's 73 `check:*` / `audit:*` scripts are referenced by no other npm script.**--## Binding operating clause--1. Finish 100% of what this machine can do. Never end with a question or an unexecuted plan.-2. **Do not wire a red guard into `gate`.** `gate` is the deploy path for every agent here;- adding a failing check blocks everyone's work, including concurrent sessions. A guard is- either fixed to green and then wired, or left unwired with its reason recorded.-3. Hard rules: no mocks, no shortcuts, explicit-path commits only, no em-dash characters.--## Step 0: re-derive the state (trust no number in this file)--```bash-node -e "-const s=require('./package.json').scripts, all=Object.keys(s);-const bodies=all.map(k=>s[k]).join(' && ');-const guards=all.filter(k=>/^(check|audit):/.test(k));-const unwired=guards.filter(k=>!new RegExp('npm run '+k.replace(/[:.]/g,'\\\\\$&')+'(\\\\s|\$|&)').test(bodies));-console.log(guards.length+' guards, '+unwired.length+' unwired'); unwired.forEach(u=>console.log(' '+u));-"-```--Measured 2026-09-04: 73 guards, 39 unwired. Sampling ten of the cheap repo-only ones gave:--| Guard | Exit | Note |-|---|---|---|-| `audit:deploy` | 0 | Already covered by `tests/deploy-artifacts.test.js`, so `npm test` runs it. Wire or leave, but say which. |-| `check:skills-seed` | 0 | Green, cheap, wireable today. |-| `audit:motion` | 0 | Green, cheap, wireable today. |-| `audit:route-shadowing` | 1 | Real findings. Fix first. |-| `check:runnable-docs` | 1 | Real findings. Fix first. |-| `check:doc-media` | 1 | 30 problems. Fix first. |-| `check:docs-search` | 1 | Committed index stale; `npm run build:docs-search` and commit. |-| `audit:tour-global` | 1 | `public/tour-builder/tour.global.js` stale; `npm run build:tour-global`. |-| `audit:csp` | 124 | Exceeded 150s. Measure its real runtime before considering it for `gate`. |-| `audit:overlays` | 124 | Same. |--The other 29 were not sampled. Many are legitimately manual: they need a browser-(`audit:web*`, `audit:a11y`), live network or funds (`audit:service-wallets`,-`check:relayer-balances`, `check:evm-rpc`), a credential (`audit:custodial-keys`), or live-`gcloud` (`check:cron-drift`, which belongs to-[fix-queue-03](905-fix-queue-03-cron-drift-garment-sweep.md)). `check:rules` and `check:secrets`-are wired into the pre-push hook rather than a script, which counts as wired. Classify before-you judge.--## Tasks--1. **Classify all 39.** Produce a table in `docs/ops/guard-wiring.md` with one row per unwired- guard: what it checks, its measured exit code and runtime, and a verdict of `gate`,- `npm test`, `deploy path`, `manual (reason)`, or `delete (superseded)`. Measure every row;- do not infer a verdict from the script's name.-2. **Fix the cheap reds**, in their own topical commits: the stale generated artifacts- (`check:docs-search`, `audit:tour-global`) are a regenerate-and-commit; `check:doc-media`,- `check:runnable-docs` and `audit:route-shadowing` need their findings read and fixed.-3. **Wire the ones that earn it.** Green plus fast (under about 15s) plus repo-only goes into- `gate`. Anything needing network, a browser, or a credential does not go into `gate`; give- it a `npm test` home or leave it manual with the reason recorded in the doc.-4. **Pin each new wiring with a test**, following `tests/gcloudignore-wiring.test.js`: assert- the guard is referenced by the script that is supposed to run it, so the next `git add -A`- sweep cannot silently unwire it.-5. **Delete what is superseded.** A guard whose job is fully covered by a vitest suite is dead- weight; remove the script and say so in the doc rather than leaving a decoy.--## Definition of done--- [ ] `docs/ops/guard-wiring.md` exists, is linked from `docs/ops/README.md`, and has a- measured row for every guard the step 0 command reports as unwired.-- [ ] Every guard the doc marks `gate` is actually in `gate`, and `npm run gate` exits 0.-- [ ] Every newly wired guard has a wiring assertion in a test file, and `npm test` passes.-- [ ] The step 0 command's unwired count has dropped, and every guard still on the list has a- recorded reason in the doc.-- [ ] `npm run audit:docs` exits 0; `npm run check:rules -- --paths <files you touched>` clean.-- [ ] Run logged in [PROGRESS.md](_context/production-100-PROGRESS.md).--## Never blocked--| Blocker | Resolution |-|---|---|-| A guard is red and the fix is large | Do not wire it. Record the finding count in the doc and open a numbered order for the fix. A recorded red beats a hidden one. |-| A guard needs a credential this box lacks | It is `manual` by definition. Record the credential it needs and which OWNER-ACTIONS row covers it, if any. |-| `gate` is already red from another agent's work | Capture the baseline first, then judge your own additions against that baseline, never against green. |-| A guard times out | Measure it with a longer timeout before judging. A slow guard is a `npm test` or deploy-path candidate, not a `gate` one. |--## Report format--State the before and after unwired counts, the guards you wired and where, the reds you fixed-against the ones you recorded, and anything you deleted with the reason.diff --git a/prompts/finish/014-openai-pr-06-docs-accuracy-reconciliation.md b/prompts/finish/014-openai-pr-06-docs-accuracy-reconciliation.mddeleted file mode 100644index d7c5692f2..000000000--- a/prompts/finish/014-openai-pr-06-docs-accuracy-reconciliation.md+++ /dev/null@@ -1,94 +0,0 @@-# Task 06: Fix doc accuracy and reconcile the tool-count story (P1)--Read [`openai-pr-00-START-HERE.md`](_context/openai-pr-00-START-HERE.md) first. This task cleans up the-documentation drift the audit found, EXCEPT the two docs owned by other tasks-(`apps-sdk/README.md` was Task 01, already landed on 2026-07-18; the `forge_free` tier-claim is Task 03). Do not fight those tasks for those specific edits; coordinate.--## The problems (verified)--1. **Tool-count inconsistency.** The free MCP connector exposes **9** tools: 6- generation (`forge_free`, `text_to_avatar`, `mesh_forge`, `rig_mesh`,- `forge_avatar`, `refine_model`) + 3 persona (`create_agent_persona`,- `get_agent_persona`, `persona_say`). But docs tell different stories:- - `openai-submission.md` says "Nine tools."- - `docs/mcp-studio.md` headline tool table lists six, with the persona tools in a- later "Embodiment" section.- - Code comments drift: `tools.js:1-8` says "Six generation tools (five generators +- refine_model)"; `tests/mcp-studio.test.js` comment says "five allowed generation- tools" while the asserted array has six.- Reconcile these so every count is consistent and correct. The authoritative number- is whatever the running server returns from `tools/list`; verify it, then make all- docs and comments match.--2. **model-viewer version is not single-sourced.** Three surfaces load different- `model-viewer` versions: `api/ar.js` launch page (3.5.0 via jsdelivr), the MCP- widget in `component.js` (3.5.0), and `/viewer.html` (4.0.0). Pick one version,- pin it in one place they all reference (a constant/shared import), or at minimum- document the intended version per surface and why they differ. A reviewer opening- two of our surfaces should not get two different renderer versions with no reason.--3. **`docs/mcp-studio.md` cross-reference check.** It points the custom-GPT contract to- `./api-reference.md`. Verify that file exists and actually documents `/api/3d/studio`.- If not, fix the pointer to the authoritative schema (the served OpenAPI from Task 05)- or add the missing section. No dead doc links.--4. **General accuracy sweep of the ChatGPT-surface docs.** Read- `docs/mcp-studio.md`, `docs/chatgpt-ar.md`, `docs/mcp-3d-studio.md`, and- `docs/ar.md` end to end against the code. Every `ui://` URI, file path, endpoint,- env var, and example must resolve and be correct. Fix any that are stale. The audit- already confirmed `docs/chatgpt-ar.md` correctly points at `component.js` (good) and- that `apps-sdk/README.md` was the worst offender; Task 01 rewrote it on 2026-07-18,- so treat it as correct unless a link in it stops resolving.--## Constraints--- Every rule in `openai-pr-00-START-HERE.md`. Docs are real implementations: every code sample- must run, every link must resolve to a live path (`CLAUDE.md` documentation rules).-- Not a crypto surface; commit gate does not apply. Do not introduce any non-$THREE- coin reference into the docs.-- Coordinate: `apps-sdk/README.md` was rewritten by Task 01 (landed) and is accurate;- leave the `forge_free` tier language to Task 03. If Task 03 has not landed yet, note- the dependency rather than duplicating its edits.--## Verification--- Start from the live `tools/list` output (call `/api/mcp-studio` with an- `initialize` + `tools/list` JSON-RPC batch) and confirm the count and names, then- grep every doc/comment for tool counts and confirm they match.-- `grep -rn "model-viewer@" api public docs` shows a single intended version story.-- Every internal doc link resolves (spot-check with a link checker or by opening each- referenced path).-- `npm test` green (docs changes should not break tests, but run it).--## Definition of done--- [ ] Tool count is consistent and correct (9) across `openai-submission.md`,- `docs/mcp-studio.md`, code comments, and test comments.-- [ ] model-viewer version is single-sourced or its per-surface difference is- documented and justified.-- [ ] `docs/mcp-studio.md` cross-references resolve; no dead links in any- ChatGPT-surface doc.-- [ ] All ChatGPT-surface docs are accurate against the code (paths, URIs, endpoints,- env vars, examples).-- [ ] `data/changelog.json` entry with the `docs` tag.-- [ ] `npm test` green.--## Retire this prompt when it is done (required)--1. Verify every Definition of done line against actual command output in front- of you. Never claim a line you did not verify.-2. Record the outcome in this campaign's PROGRESS or INDEX file if it has one.-3. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares), and delete this- prompt file in that same commit:-- git rm prompts/finish/014-openai-pr-06-docs-accuracy-reconciliation.md-- A finished order left on disk reads as open work to the next agent, so the- shrinking directory is the campaign's progress ledger.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns it.-Never delete this file on a partial.diff --git a/prompts/finish/105-swarm-100-route-temporary.md b/prompts/finish/105-swarm-100-route-temporary.mddeleted file mode 100644index 308b54bf0..000000000--- a/prompts/finish/105-swarm-100-route-temporary.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /temporary (Drive Your Avatar)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/temporary')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/temporary" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/temporary--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /temporary ("Drive Your Avatar", section "main").-Declared purpose: Drive a 3D avatar around with a joystick or WASD, then toggle the back camera for AR-style passthrough on your phone.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/temporary. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /temporary: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/105-swarm-100-route-temporary.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/107-swarm-100-route-launches.md b/prompts/finish/107-swarm-100-route-launches.mddeleted file mode 100644index 42c9042c9..000000000--- a/prompts/finish/107-swarm-100-route-launches.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /launches (Agent Launches)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/launches')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/launches" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/launches--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /launches ("Agent Launches", section "crypto").-Declared purpose: Live public feed of every coin launched by a three.ws agent: market caps, graduation status, and the agent behind each launch.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/launches. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /launches: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/107-swarm-100-route-launches.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/110-swarm-100-route-ibm-hello.md b/prompts/finish/110-swarm-100-route-ibm-hello.mddeleted file mode 100644index d6f4023cc..000000000--- a/prompts/finish/110-swarm-100-route-ibm-hello.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /ibm/hello (IBM × three.ws partnership)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/ibm/hello')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/ibm/hello" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/ibm/hello--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /ibm/hello ("IBM × three.ws partnership", section "crypto").-Declared purpose: The IBM × three.ws partnership page: how three.ws's agent identity, x402 payment rails, and MCP servers integrate with IBM watsonx and the IBM x402 ecosystem. A self-contained showcase that stays live-editable after publish.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/ibm/hello. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /ibm/hello: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/110-swarm-100-route-ibm-hello.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/111-swarm-100-route-showcase.md b/prompts/finish/111-swarm-100-route-showcase.mddeleted file mode 100644index a538de838..000000000--- a/prompts/finish/111-swarm-100-route-showcase.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /showcase (Showcase)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/showcase')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/showcase" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/showcase--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /showcase ("Showcase", section "main").-Declared purpose: AI-powered 3D model viewer showcase: drop in a model and explore it live.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/showcase. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /showcase: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/111-swarm-100-route-showcase.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/112-swarm-100-route-home.md b/prompts/finish/112-swarm-100-route-home.mddeleted file mode 100644index 10e48d047..000000000--- a/prompts/finish/112-swarm-100-route-home.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: / (Home)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: / ("Home", section "main").-Declared purpose: Landing page. 3D AI agents that think on IBM watsonx.ai (Granite): the pitch, live agent demos, and the front door to every flow.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/112-swarm-100-route-home.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/113-swarm-100-route-marketplace.md b/prompts/finish/113-swarm-100-route-marketplace.mddeleted file mode 100644index bfde22ec8..000000000--- a/prompts/finish/113-swarm-100-route-marketplace.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /marketplace (Marketplace)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/marketplace')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/marketplace" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/marketplace--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /marketplace ("Marketplace", section "main").-Declared purpose: Buy access to agents, skills, and avatars from other creators.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/marketplace. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /marketplace: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/113-swarm-100-route-marketplace.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/114-swarm-100-route-play.md b/prompts/finish/114-swarm-100-route-play.mddeleted file mode 100644index fe9809c7a..000000000--- a/prompts/finish/114-swarm-100-route-play.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /play (Play)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/play')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/play" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/play--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /play ("Play", section "crypto").-Declared purpose: Jump into a live coin community world and explore it with your avatar.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/play. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /play: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/114-swarm-100-route-play.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/115-swarm-100-route-forge-studio.md b/prompts/finish/115-swarm-100-route-forge-studio.mddeleted file mode 100644index b6357b400..000000000--- a/prompts/finish/115-swarm-100-route-forge-studio.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /forge-studio (Studio: Object + Avatar from text)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/forge-studio')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/forge-studio" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/forge-studio--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /forge-studio ("Studio: Object + Avatar from text", section "build").-Declared purpose: One canvas for both text-to-3D pipelines: forge a textured object (GLB) in the Object tab, or describe a character and get a rigged avatar in the Avatar tab.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/forge-studio. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /forge-studio: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/115-swarm-100-route-forge-studio.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/116-swarm-100-route-pricing.md b/prompts/finish/116-swarm-100-route-pricing.mddeleted file mode 100644index 23458c2b3..000000000--- a/prompts/finish/116-swarm-100-route-pricing.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /pricing (Pricing)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/pricing')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/pricing" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/pricing--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /pricing ("Pricing", section "main").-Declared purpose: Plans and pricing for three.ws: free forge quota, premium passes, and pay-per-use API surfaces.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/pricing. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /pricing: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/116-swarm-100-route-pricing.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/117-swarm-100-route-login.md b/prompts/finish/117-swarm-100-route-login.mddeleted file mode 100644index 06e9415a5..000000000--- a/prompts/finish/117-swarm-100-route-login.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /login (Sign in)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/login')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/login" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/login--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /login ("Sign in", section "account").-Declared purpose: Sign in with email, GitHub, or a connected wallet.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/login. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /login: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/117-swarm-100-route-login.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/118-swarm-100-route-register.md b/prompts/finish/118-swarm-100-route-register.mddeleted file mode 100644index 2b8db5cdc..000000000--- a/prompts/finish/118-swarm-100-route-register.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /register (Create account)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/register')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/register" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/register--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /register ("Create account", section "account").-Declared purpose: Create a new three.ws account.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/register. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /register: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/118-swarm-100-route-register.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/119-swarm-100-route-search.md b/prompts/finish/119-swarm-100-route-search.mddeleted file mode 100644index 129526cf2..000000000--- a/prompts/finish/119-swarm-100-route-search.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /search (Search)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/search')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/search" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/search--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /search ("Search", section "main").-Declared purpose: One search across every creation type on three.ws: avatars, agents, 3D models, worlds, and coins: with remixable models remixable straight from the results.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/search. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /search: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/119-swarm-100-route-search.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/120-swarm-100-route-settings.md b/prompts/finish/120-swarm-100-route-settings.mddeleted file mode 100644index ba79db98c..000000000--- a/prompts/finish/120-swarm-100-route-settings.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /settings (Settings)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/settings')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/settings" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/settings--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /settings ("Settings", section "account").-Declared purpose: Account, profile, and notification settings.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/settings. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /settings: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/120-swarm-100-route-settings.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/121-swarm-100-route-profile.md b/prompts/finish/121-swarm-100-route-profile.mddeleted file mode 100644index 8e3fafe3d..000000000--- a/prompts/finish/121-swarm-100-route-profile.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /profile (Your creator portfolio)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/profile')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/profile" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/profile--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /profile ("Your creator portfolio", section "account").-Declared purpose: Your public three.ws portfolio: every avatar, agent, forged 3D model, world, coin, skill, and plugin you've made, in one place. Signed out, this redirects to sign-in; every user's own page also lives at /u/:username.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/profile. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /profile: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/121-swarm-100-route-profile.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/122-swarm-100-route-dad.md b/prompts/finish/122-swarm-100-route-dad.mddeleted file mode 100644index 9cc205d06..000000000--- a/prompts/finish/122-swarm-100-route-dad.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /dad (Make Dad a 3D Avatar)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/dad')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/dad" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/dad--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /dad ("Make Dad a 3D Avatar", section "build").-Declared purpose: Turn one photo of your dad into a recognizable, rigged, animated 3D avatar: it idles, walks, and waves: then share a permalink.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/dad. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /dad: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/122-swarm-100-route-dad.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/123-swarm-100-route-daily-match.md b/prompts/finish/123-swarm-100-route-daily-match.mddeleted file mode 100644index d1072fac3..000000000--- a/prompts/finish/123-swarm-100-route-daily-match.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /daily-match (Daily Match)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/daily-match')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/daily-match" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/daily-match--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /daily-match ("Daily Match", section "labs").-Declared purpose: Which agent ships the most real output today? Live daily standings over real agent activity: actions, trades, skill sales, and coin launches. Resets at 00:00 UTC.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/daily-match. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /daily-match: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/123-swarm-100-route-daily-match.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/124-swarm-100-route-dashboard-settings.md b/prompts/finish/124-swarm-100-route-dashboard-settings.mddeleted file mode 100644index 3efc20b6f..000000000--- a/prompts/finish/124-swarm-100-route-dashboard-settings.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /dashboard/settings (Dashboard · Settings)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/dashboard/settings')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/dashboard/settings" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/dashboard/settings--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /dashboard/settings ("Dashboard · Settings", section "account").-Declared purpose: Avatar storage modes (R2 vs IPFS), sessions, your notification inbox, per-category notification channel preferences (in-app/push/email/Telegram), and dashboard preferences.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/dashboard/settings. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /dashboard/settings: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/124-swarm-100-route-dashboard-settings.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/125-swarm-100-route-demo.md b/prompts/finish/125-swarm-100-route-demo.mddeleted file mode 100644index 3f85e56a8..000000000--- a/prompts/finish/125-swarm-100-route-demo.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /demo (Agent Economy: NOVA & ORACLE)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/demo')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/demo" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/demo--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /demo ("Agent Economy: NOVA & ORACLE", section "crypto").-Declared purpose: Two three.ws AI agents with real Solana wallets discover services on the x402 bazaar, pay each other in SOL, and display a live market briefing on the TV in their world.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/demo. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /demo: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/125-swarm-100-route-demo.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/126-swarm-100-route-demos.md b/prompts/finish/126-swarm-100-route-demos.mddeleted file mode 100644index 1ea24a6c2..000000000--- a/prompts/finish/126-swarm-100-route-demos.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /demos (Lab Demos)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/demos')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/demos" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/demos--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /demos ("Lab Demos", section "main").-Declared purpose: Index of live demos across the platform: viewers, agents, payments, and experiments.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/demos. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /demos: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/126-swarm-100-route-demos.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/127-swarm-100-route-deployments.md b/prompts/finish/127-swarm-100-route-deployments.mddeleted file mode 100644index 6ed71feff..000000000--- a/prompts/finish/127-swarm-100-route-deployments.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /deployments (On-chain Deployments)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/deployments')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/deployments" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/deployments--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /deployments ("On-chain Deployments", section "crypto").-Declared purpose: A live, cross-chain feed of every agent registered on the ERC-8004 Identity Registry: the moment each registration lands on-chain. Real on-chain registrations only, with chain footprint, 3D-avatar and x402 capability stats.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/deployments. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /deployments: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/127-swarm-100-route-deployments.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/128-swarm-100-route-derivatives.md b/prompts/finish/128-swarm-100-route-derivatives.mddeleted file mode 100644index ebff951e5..000000000--- a/prompts/finish/128-swarm-100-route-derivatives.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /derivatives (Crypto Derivatives & Perpetual Futures)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/derivatives')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/derivatives" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/derivatives--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /derivatives ("Crypto Derivatives & Perpetual Futures", section "crypto").-Declared purpose: Live perpetual-futures markets across exchanges: price, 24h change, funding rate, open interest, and volume for every major contract. Filter by underlying index (BTC, ETH, SOL) and sort by any column.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/derivatives. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /derivatives: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/128-swarm-100-route-derivatives.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/129-swarm-100-route-dex-volumes.md b/prompts/finish/129-swarm-100-route-dex-volumes.mddeleted file mode 100644index 9fa8fb600..000000000--- a/prompts/finish/129-swarm-100-route-dex-volumes.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /dex-volumes (DEX Volume Leaderboard)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/dex-volumes')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/dex-volumes" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/dex-volumes--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /dex-volumes ("DEX Volume Leaderboard", section "crypto").-Declared purpose: Every decentralized exchange ranked by trading volume, with 24h and 7d totals, 7-day change, and market share, plus an aggregate volume history chart. Live from DeFiLlama.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/dex-volumes. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /dex-volumes: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/129-swarm-100-route-dex-volumes.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/130-swarm-100-route-diorama.md b/prompts/finish/130-swarm-100-route-diorama.mddeleted file mode 100644index ee1928d34..000000000--- a/prompts/finish/130-swarm-100-route-diorama.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /diorama (Diorama)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/diorama')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/diorama" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/diorama--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /diorama ("Diorama", section "main").-Declared purpose: Speak a little world into being: describe a scene and watch it assemble into an explorable 3D diorama you can walk through, then take it into AR.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/diorama. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /diorama: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/130-swarm-100-route-diorama.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/131-swarm-100-route-discover.md b/prompts/finish/131-swarm-100-route-discover.mddeleted file mode 100644index 05ea7be92..000000000--- a/prompts/finish/131-swarm-100-route-discover.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /discover (Discover)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/discover')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/discover" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/discover--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /discover ("Discover", section "main").-Declared purpose: On-chain agent directory (ERC-8004 + Solana). Browse, search, and inspect every registered agent.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/discover. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /discover: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/131-swarm-100-route-discover.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/132-swarm-100-route-embed-doctor.md b/prompts/finish/132-swarm-100-route-embed-doctor.mddeleted file mode 100644index 47388910b..000000000--- a/prompts/finish/132-swarm-100-route-embed-doctor.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /embed-doctor (Embed Doctor)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/embed-doctor')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/embed-doctor" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/embed-doctor--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /embed-doctor ("Embed Doctor", section "build").-Declared purpose: Find out why your 3D agent embed is not showing up. Paste your page URL and we load it in a real browser, watch the embed try to boot, and name the exact line to change: a Content-Security-Policy that dropped the runtime, a container with no height, a private agent id, WebGL turned off. Works on a snippet too, before you deploy it.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/embed-doctor. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /embed-doctor: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/132-swarm-100-route-embed-doctor.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/133-swarm-100-route-eth-vanity.md b/prompts/finish/133-swarm-100-route-eth-vanity.mddeleted file mode 100644index 59e2d0942..000000000--- a/prompts/finish/133-swarm-100-route-eth-vanity.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /eth-vanity (ETH Vanity (CREATE2))--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/eth-vanity')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/eth-vanity" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/eth-vanity--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /eth-vanity ("ETH Vanity (CREATE2)", section "crypto").-Declared purpose: Grind a vanity smart-contract address using the ThreeWSFactory CREATE2 deployer on BSC.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/eth-vanity. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /eth-vanity: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/133-swarm-100-route-eth-vanity.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/134-swarm-100-route-event.md b/prompts/finish/134-swarm-100-route-event.mddeleted file mode 100644index 926d89b55..000000000--- a/prompts/finish/134-swarm-100-route-event.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /event (Live event)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/event')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/event" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/event--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /event ("Live event", section "crypto").-Declared purpose: The live three.ws community meetup: a ticking countdown off the same config /play reads, the run of show in your own timezone, an add-to-calendar file, who is in the world right now, and the one door in.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/event. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /event: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/134-swarm-100-route-event.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/135-swarm-100-route-evm-wallet.md b/prompts/finish/135-swarm-100-route-evm-wallet.mddeleted file mode 100644index f25e332a5..000000000--- a/prompts/finish/135-swarm-100-route-evm-wallet.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /evm-wallet (EVM Vanity Wallet)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/evm-wallet')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/evm-wallet" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/evm-wallet--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /evm-wallet ("EVM Vanity Wallet", section "crypto").-Declared purpose: Grind a custom Ethereum / EVM wallet address with the prefix or suffix of your choice. A real secp256k1 keypair, generated entirely in your browser: the private key never leaves your device.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/evm-wallet. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /evm-wallet: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/135-swarm-100-route-evm-wallet.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/136-swarm-100-route-exchanges.md b/prompts/finish/136-swarm-100-route-exchanges.mddeleted file mode 100644index 81a53a7c2..000000000--- a/prompts/finish/136-swarm-100-route-exchanges.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /exchanges (Top Crypto Exchanges)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/exchanges')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/exchanges" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/exchanges--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /exchanges ("Top Crypto Exchanges", section "crypto").-Declared purpose: The leading cryptocurrency exchanges ranked by trust score and 24-hour trading volume, with country, launch year, and a direct link to each. Real CoinGecko data, sortable by any column.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/exchanges. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /exchanges: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/136-swarm-100-route-exchanges.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/138-swarm-100-route-fact-check.md b/prompts/finish/138-swarm-100-route-fact-check.mddeleted file mode 100644index a4af74f2e..000000000--- a/prompts/finish/138-swarm-100-route-fact-check.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /fact-check (Fact Check)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/fact-check')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/fact-check" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/fact-check--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /fact-check ("Fact Check", section "crypto").-Declared purpose: Sourced fact-checking with cryptographic attestations: a published, checkable accuracy benchmark plus a live free-check box. 3 free checks/day per IP, then $0.10 USDC via x402.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/fact-check. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /fact-check: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/138-swarm-100-route-fact-check.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/139-swarm-100-route-fact-checker.md b/prompts/finish/139-swarm-100-route-fact-checker.mddeleted file mode 100644index 6d51e1205..000000000--- a/prompts/finish/139-swarm-100-route-fact-checker.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /fact-checker (Fact Checker)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/fact-checker')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/fact-checker" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/fact-checker--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /fact-checker ("Fact Checker", section "crypto").-Declared purpose: Pay $0.10 per claim. The agent searches across authoritative sources, pays per retrieval via x402, and returns a verdict with cited evidence and a signed attestation.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/fact-checker. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /fact-checker: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/139-swarm-100-route-fact-checker.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/140-swarm-100-route-fear-greed.md b/prompts/finish/140-swarm-100-route-fear-greed.mddeleted file mode 100644index fe33a662b..000000000--- a/prompts/finish/140-swarm-100-route-fear-greed.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /fear-greed (Crypto Fear & Greed Index)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/fear-greed')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/fear-greed" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/fear-greed--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /fear-greed ("Crypto Fear & Greed Index", section "crypto").-Declared purpose: The Crypto Fear & Greed Index: live market sentiment from 0 (Extreme Fear) to 100 (Extreme Greed), shown on a gauge with a week-over-week delta, a full interactive history chart, and a labelled scale explaining what each reading means.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/fear-greed. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /fear-greed: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/140-swarm-100-route-fear-greed.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/141-swarm-100-route-features-agent-exchange.md b/prompts/finish/141-swarm-100-route-features-agent-exchange.mddeleted file mode 100644index 09db03916..000000000--- a/prompts/finish/141-swarm-100-route-features-agent-exchange.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /features/agent-exchange (Agent Exchange: AI Agents Paying Each Other)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/features/agent-exchange')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/features/agent-exchange" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/features/agent-exchange--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /features/agent-exchange ("Agent Exchange: AI Agents Paying Each Other", section "build").-Declared purpose: Watch two AI agents with 3D avatars autonomously negotiate and pay each other for crypto intelligence via x402 micropayments on Solana.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/features/agent-exchange. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /features/agent-exchange: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/141-swarm-100-route-features-agent-exchange.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/142-swarm-100-route-features-ar.md b/prompts/finish/142-swarm-100-route-features-ar.mddeleted file mode 100644index 7919f9d7a..000000000--- a/prompts/finish/142-swarm-100-route-features-ar.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /features/ar (AR & WebXR: Place Any 3D Agent in Your Real World)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/features/ar')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/features/ar" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/features/ar--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /features/ar ("AR & WebXR: Place Any 3D Agent in Your Real World", section "build").-Declared purpose: Every three.ws avatar and Forge model has a View in AR button. On mobile it anchors the model to real surfaces through your camera. WebXR on Android, Quick Look on iOS: no app, no download.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/features/ar. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /features/ar: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/142-swarm-100-route-features-ar.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/143-swarm-100-route-features-deploy.md b/prompts/finish/143-swarm-100-route-features-deploy.mddeleted file mode 100644index 93e08f81b..000000000--- a/prompts/finish/143-swarm-100-route-features-deploy.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /features/deploy (Deploy: On-Chain Agent Identity)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/features/deploy')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/features/deploy" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/features/deploy--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /features/deploy ("Deploy: On-Chain Agent Identity", section "build").-Declared purpose: Register your AI agent on Solana via ERC-8004 and Metaplex Core. Permanent, verifiable identity: discoverable by any wallet or other agent.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/features/deploy. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /features/deploy: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/143-swarm-100-route-features-deploy.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/144-swarm-100-route-features-forge.md b/prompts/finish/144-swarm-100-route-features-forge.mddeleted file mode 100644index 81324b0a6..000000000--- a/prompts/finish/144-swarm-100-route-features-forge.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /features/forge (Forge: Text & Image to 3D Models)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/features/forge')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/features/forge" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/features/forge--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /features/forge ("Forge: Text & Image to 3D Models", section "build").-Declared purpose: Type a description or upload photos, and get a downloadable textured 3D model (GLB). Multiple self-hosted engines, from fast drafts to PBR realism.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/features/forge. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /features/forge: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/144-swarm-100-route-features-forge.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/145-swarm-100-route-features-marketplace.md b/prompts/finish/145-swarm-100-route-features-marketplace.mddeleted file mode 100644index 66b7c6e66..000000000--- a/prompts/finish/145-swarm-100-route-features-marketplace.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /features/marketplace (Marketplace: Discover and Fork AI Agents)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/features/marketplace')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/features/marketplace" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/features/marketplace--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /features/marketplace ("Marketplace: Discover and Fork AI Agents", section "build").-Declared purpose: Browse hundreds of community-built AI agents with 3D avatars and on-chain identities. Fork any agent, buy paid skills, and ship in minutes.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/features/marketplace. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /features/marketplace: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/145-swarm-100-route-features-marketplace.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/146-swarm-100-route-features-play.md b/prompts/finish/146-swarm-100-route-features-play.mddeleted file mode 100644index 5fd782426..000000000--- a/prompts/finish/146-swarm-100-route-features-play.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /features/play (Play: Live 3D Coin Worlds)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/features/play')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/features/play" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/features/play--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /features/play ("Play: Live 3D Coin Worlds", section "build").-Declared purpose: Every pump.fun coin gets a deterministic 3D world. Walk in as your avatar, chat with holders, and trade: all in the browser.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/features/play. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /features/play: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/146-swarm-100-route-features-play.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/147-swarm-100-route-features-scan.md b/prompts/finish/147-swarm-100-route-features-scan.mddeleted file mode 100644index 5724c86b2..000000000--- a/prompts/finish/147-swarm-100-route-features-scan.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /features/scan (Scan: Selfie to 3D Avatar)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/features/scan')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/features/scan" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/features/scan--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /features/scan ("Scan: Selfie to 3D Avatar", section "build").-Declared purpose: Point your camera at your face, hold still, and walk away with a rigged 3D avatar in 60 seconds. Free, in your browser.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/features/scan. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /features/scan: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/147-swarm-100-route-features-scan.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/148-swarm-100-route-features-studio.md b/prompts/finish/148-swarm-100-route-features-studio.mddeleted file mode 100644index b1dc97106..000000000--- a/prompts/finish/148-swarm-100-route-features-studio.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /features/studio (Studio: Embeddable AI Widget Builder)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/features/studio')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/features/studio" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/features/studio--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /features/studio ("Studio: Embeddable AI Widget Builder", section "build").-Declared purpose: Configure avatar, voice, and knowledge base in the Widget Studio, then copy one script tag to embed a 3D AI agent anywhere.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/features/studio. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /features/studio: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/148-swarm-100-route-features-studio.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/149-swarm-100-route-features-walk.md b/prompts/finish/149-swarm-100-route-features-walk.mddeleted file mode 100644index 3d161b6d5..000000000--- a/prompts/finish/149-swarm-100-route-features-walk.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /features/walk (Walk: 3D Avatar AR)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/features/walk')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/features/walk" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/features/walk--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /features/walk ("Walk: 3D Avatar AR", section "build").-Declared purpose: Drive your AI agent's 3D avatar with WASD or joystick. Toggle AR mode and it walks on your real floor through your phone camera.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/features/walk. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /features/walk: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/149-swarm-100-route-features-walk.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/150-swarm-100-route-features.md b/prompts/finish/150-swarm-100-route-features.mddeleted file mode 100644index 17cae8154..000000000--- a/prompts/finish/150-swarm-100-route-features.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /features (Features)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/features')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/features" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/features--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /features ("Features", section "build").-Declared purpose: Full feature overview of the three.ws platform: avatars, agents, on-chain identity, x402 payments, and embedding.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/features. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /features: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/150-swarm-100-route-features.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/151-swarm-100-route-feed.md b/prompts/finish/151-swarm-100-route-feed.mddeleted file mode 100644index b22394ba6..000000000--- a/prompts/finish/151-swarm-100-route-feed.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /feed (Feed)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/feed')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/feed" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/feed--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /feed ("Feed", section "main").-Declared purpose: Live avatars, agents, 3D models, worlds, and launches across three.ws: filter to everyone or just the people you follow.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/feed. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /feed: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/151-swarm-100-route-feed.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/152-swarm-100-route-fits.md b/prompts/finish/152-swarm-100-route-fits.mddeleted file mode 100644index 13dcb4cbd..000000000--- a/prompts/finish/152-swarm-100-route-fits.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /fits (Fits: the cosmetics economy)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/fits')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/fits" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/fits--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /fits ("Fits: the cosmetics economy", section "crypto").-Declared purpose: The live cosmetics economy: the scarcest premium fits ranked by how few wallets own them, the top collectors wearing them, the creators actually getting paid in real settled USDC, and every cosmetic sale as it lands. Includes a free creator-earnings lookup for any Solana wallet, with per-cosmetic and per-coin-world breakdowns.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/fits. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /fits: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/152-swarm-100-route-fits.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/153-swarm-100-route-forever.md b/prompts/finish/153-swarm-100-route-forever.mddeleted file mode 100644index f84e22b1b..000000000--- a/prompts/finish/153-swarm-100-route-forever.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /forever (Forever: etch a message into Bitcoin)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/forever')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/forever" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/forever--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /forever ("Forever: etch a message into Bitcoin", section "crypto").-Declared purpose: Inscribe a message onto the Bitcoin blockchain. It stays there. Forever.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/forever. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /forever: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/153-swarm-100-route-forever.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/154-swarm-100-route-gallery.md b/prompts/finish/154-swarm-100-route-gallery.mddeleted file mode 100644index bd63f6b26..000000000--- a/prompts/finish/154-swarm-100-route-gallery.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /gallery (Avatar Gallery)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/gallery')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/gallery" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/gallery--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /gallery ("Avatar Gallery", section "main").-Declared purpose: Every public 3D avatar in the system, browsable as a grid.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/gallery. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /gallery: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/154-swarm-100-route-gallery.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/155-swarm-100-route-gas.md b/prompts/finish/155-swarm-100-route-gas.mddeleted file mode 100644index 0fd8c3c7a..000000000--- a/prompts/finish/155-swarm-100-route-gas.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /gas (Ethereum Gas Tracker)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/gas')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/gas" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/gas--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /gas ("Ethereum Gas Tracker", section "crypto").-Declared purpose: Live Ethereum gas fees straight from the chain: slow, standard, and fast tiers in gwei with USD cost estimates for transfers, token sends, DEX swaps, and NFT mints. Real on-chain fee history, refreshed automatically, no API key required.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/gas. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /gas: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/155-swarm-100-route-gas.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/156-swarm-100-route-genesis.md b/prompts/finish/156-swarm-100-route-genesis.mddeleted file mode 100644index 52b322852..000000000--- a/prompts/finish/156-swarm-100-route-genesis.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /genesis (Instant Agent Genesis)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/genesis')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/genesis" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/genesis--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /genesis ("Instant Agent Genesis", section "build").-Declared purpose: A prompt or a selfie becomes a rigged 3D AI agent with its own custodial Solana + EVM wallet and a verifiable on-chain identity: in under a minute, owned by you.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/genesis. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /genesis: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/156-swarm-100-route-genesis.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/157-swarm-100-route-genome.md b/prompts/finish/157-swarm-100-route-genome.mddeleted file mode 100644index 45943aefb..000000000--- a/prompts/finish/157-swarm-100-route-genome.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /genome (Agent Genome: breed two agents)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/genome')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/genome" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/genome--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /genome ("Agent Genome: breed two agents", section "build").-Declared purpose: Breed two AI agents into a provably-inherited offspring. The child inherits a recombination of both parents' brain, voice, body, and on-chain skill licenses: each breeding event seed-recorded so the lineage is re-derivable and forgery-detectable, with a verifiable family tree on every parent and child.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/genome. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /genome: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/157-swarm-100-route-genome.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/158-swarm-100-route-gestures.md b/prompts/finish/158-swarm-100-route-gestures.mddeleted file mode 100644index 5616be3f9..000000000--- a/prompts/finish/158-swarm-100-route-gestures.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /gestures (Agent Gestures)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/gestures')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/gestures" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/gestures--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /gestures ("Agent Gestures", section "main").-Declared purpose: The three.ws agent gesture vocabulary: every animation slot, the clip it plays, the skills that trigger it, and a builder for overriding any slot on your own agent. Previewed live on a real avatar.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/gestures. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /gestures: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/158-swarm-100-route-gestures.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/159-swarm-100-route-ghost-copy.md b/prompts/finish/159-swarm-100-route-ghost-copy.mddeleted file mode 100644index d430fe7e4..000000000--- a/prompts/finish/159-swarm-100-route-ghost-copy.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /ghost-copy (Ghost-copy: what if you'd copied them?)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/ghost-copy')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/ghost-copy" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/ghost-copy--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /ghost-copy ("Ghost-copy: what if you'd copied them?", section "crypto").-Declared purpose: Paper-copy any verified pump.fun trader agent with fake money over their real closed on-chain trades. Pick a leader, a budget and a window, and see the equity curve you would have had, plus every trade your budget could not reach and why. No wallet, no signature, no custody, no account.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/ghost-copy. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /ghost-copy: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/159-swarm-100-route-ghost-copy.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/160-swarm-100-route-gmgn.md b/prompts/finish/160-swarm-100-route-gmgn.mddeleted file mode 100644index 9810d8010..000000000--- a/prompts/finish/160-swarm-100-route-gmgn.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /gmgn (GMGN Smart Money)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/gmgn')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/gmgn" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/gmgn--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /gmgn ("GMGN Smart Money", section "crypto").-Declared purpose: Live smart-money signals narrated by a 3D AI agent. Tracks which wallets are loading on Solana, Ethereum, Base, and BNB Chain in real time.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/gmgn. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /gmgn: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/160-swarm-100-route-gmgn.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/161-swarm-100-route-hacks.md b/prompts/finish/161-swarm-100-route-hacks.mddeleted file mode 100644index 2fa4b9458..000000000--- a/prompts/finish/161-swarm-100-route-hacks.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /hacks (DeFi Hacks Database)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/hacks')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/hacks" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/hacks--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /hacks ("DeFi Hacks Database", section "crypto").-Declared purpose: A searchable database of every major DeFi exploit: amount stolen, classification, technique, chains affected, whether it was a bridge hack, and the source, with all-time and trailing-12-month totals. Live from DeFiLlama.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/hacks. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /hacks: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/161-swarm-100-route-hacks.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/162-swarm-100-route-heatmap.md b/prompts/finish/162-swarm-100-route-heatmap.mddeleted file mode 100644index 4b46c59d0..000000000--- a/prompts/finish/162-swarm-100-route-heatmap.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /heatmap (Crypto Market Heatmap)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/heatmap')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/heatmap" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/heatmap--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /heatmap ("Crypto Market Heatmap", section "crypto").-Declared purpose: A live market heatmap of the top cryptocurrencies: every tile is a coin, sized by market cap and colored by its 24h or 7d price move. See where money is flowing across the whole market at a glance, then click any tile for its full profile.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/heatmap. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /heatmap: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/162-swarm-100-route-heatmap.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/163-swarm-100-route-hero-demo.md b/prompts/finish/163-swarm-100-route-hero-demo.mddeleted file mode 100644index cc3eabb85..000000000--- a/prompts/finish/163-swarm-100-route-hero-demo.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /hero-demo (Hero Stage)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/hero-demo')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/hero-demo" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/hero-demo--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /hero-demo ("Hero Stage", section "crypto").-Declared purpose: A cinematic 3D hero stage for three.ws agents: a lit avatar centered in deep space over concentric glowing rings, a drifting starfield, and soft bloom, with pointer parallax and a live avatar switcher. Motion-safe.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/hero-demo. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /hero-demo: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/163-swarm-100-route-hero-demo.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/164-swarm-100-route-holo.md b/prompts/finish/164-swarm-100-route-holo.mddeleted file mode 100644index 68b6a0fdd..000000000--- a/prompts/finish/164-swarm-100-route-holo.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /holo (Holo Sticker)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/holo')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/holo" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/holo--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /holo ("Holo Sticker", section "labs").-Declared purpose: A live holographic foil sticker rendered in your browser: tilt it with your pointer, peel the corner, switch foils, die-cut your own text, and download the frame as a PNG. Fully procedural, no assets.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/holo. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /holo: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/164-swarm-100-route-holo.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/165-swarm-100-route-hydrate.md b/prompts/finish/165-swarm-100-route-hydrate.mddeleted file mode 100644index fc1926f3d..000000000--- a/prompts/finish/165-swarm-100-route-hydrate.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /hydrate (Hydrate)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/hydrate')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/hydrate" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/hydrate--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /hydrate ("Hydrate", section "build").-Declared purpose: Take an existing on-chain agent (ERC-8004 / Solana) and attach a 3D body, voice, and skills.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/hydrate. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /hydrate: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/165-swarm-100-route-hydrate.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/166-swarm-100-route-ibm-x402-demo.md b/prompts/finish/166-swarm-100-route-ibm-x402-demo.mddeleted file mode 100644index 5345087dd..000000000--- a/prompts/finish/166-swarm-100-route-ibm-x402-demo.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /ibm/x402-demo (x402 Live Demo (IBM × three.ws))--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/ibm/x402-demo')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/ibm/x402-demo" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/ibm/x402-demo--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /ibm/x402-demo ("x402 Live Demo (IBM × three.ws)", section "crypto").-Declared purpose: A self-contained x402 micropayment demo: pay $0.001 USDC from your own wallet: MetaMask on Base or Phantom on Solana: to call a real three.ws data API. Watch the live HTTP 402 challenge, sign a gasless authorization, and see the on-chain settlement receipt. No backend, no account; built as the public face of the IBM × three.ws partnership.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/ibm/x402-demo. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /ibm/x402-demo: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/166-swarm-100-route-ibm-x402-demo.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/167-swarm-100-route-image-to-3d.md b/prompts/finish/167-swarm-100-route-image-to-3d.mddeleted file mode 100644index f4d675398..000000000--- a/prompts/finish/167-swarm-100-route-image-to-3d.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /image-to-3d (Image to 3D: Turn a Photo into a 3D Model)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/image-to-3d')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/image-to-3d" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/image-to-3d--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /image-to-3d ("Image to 3D: Turn a Photo into a 3D Model", section "build").-Declared purpose: Upload a photo (or up to 4 angles) and get a downloadable textured 3D model (GLB). The Forge's photo lane, opened directly: for objects and props; for faces, use Selfie to Avatar.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/image-to-3d. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /image-to-3d: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/167-swarm-100-route-image-to-3d.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/168-swarm-100-route-import-rpm.md b/prompts/finish/168-swarm-100-route-import-rpm.mddeleted file mode 100644index 19512fea2..000000000--- a/prompts/finish/168-swarm-100-route-import-rpm.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /import/rpm (Import an avatar URL)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/import/rpm')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/import/rpm" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/import/rpm--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /import/rpm ("Import an avatar URL", section "build").-Declared purpose: Import any GLB or glTF avatar into three.ws and give it an agent brain.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/import/rpm. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /import/rpm: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/168-swarm-100-route-import-rpm.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/169-swarm-100-route-inspect.md b/prompts/finish/169-swarm-100-route-inspect.mddeleted file mode 100644index 86eb5020d..000000000--- a/prompts/finish/169-swarm-100-route-inspect.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /inspect (Model Inspector)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/inspect')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/inspect" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/inspect--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /inspect ("Model Inspector", section "build").-Declared purpose: Run the exact quality gate three.ws applies to its own catalog against any .glb, in your browser. Nothing is uploaded. Per-rule verdict, what was measured against each bound, and the fix for every failure.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/inspect. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /inspect: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/169-swarm-100-route-inspect.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/170-swarm-100-route-integrations.md b/prompts/finish/170-swarm-100-route-integrations.mddeleted file mode 100644index f139cc14e..000000000--- a/prompts/finish/170-swarm-100-route-integrations.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /integrations (Integrations)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/integrations')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/integrations" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/integrations--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /integrations ("Integrations", section "build").-Declared purpose: Drop-in 3D agents, chat widgets, walk companions, and live token embeds for any site. One script tag. Powered by three.ws.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/integrations. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /integrations: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/170-swarm-100-route-integrations.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/171-swarm-100-route-irl-privacy.md b/prompts/finish/171-swarm-100-route-irl-privacy.mddeleted file mode 100644index 960fde695..000000000--- a/prompts/finish/171-swarm-100-route-irl-privacy.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /irl-privacy (How location works on IRL)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/irl-privacy')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/irl-privacy" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/irl-privacy--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /irl-privacy ("How location works on IRL", section "main").-Declared purpose: Placed agents are private by location: they appear only to people physically standing near them, never on a map or a list. What's shared, what isn't, and how your camera, motion, and location are used.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/irl-privacy. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /irl-privacy: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/171-swarm-100-route-irl-privacy.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/172-swarm-100-route-irl.md b/prompts/finish/172-swarm-100-route-irl.mddeleted file mode 100644index a09753420..000000000--- a/prompts/finish/172-swarm-100-route-irl.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /irl (IRL)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/irl')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/irl" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/irl--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /irl ("IRL", section "main").-Declared purpose: Place a 3D avatar in your real environment. Camera AR passthrough, joystick movement, and tap-to-place 3D objects on your floor.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/irl. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /irl: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/172-swarm-100-route-irl.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/173-swarm-100-route-labor-market.md b/prompts/finish/173-swarm-100-route-labor-market.mddeleted file mode 100644index e59d1f7fc..000000000--- a/prompts/finish/173-swarm-100-route-labor-market.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /labor-market (Agent Labor Market)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/labor-market')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/labor-market" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/labor-market--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /labor-market ("Agent Labor Market", section "crypto").-Declared purpose: A live machine economy: agents post bounties, bid on each other's work, and settle in $THREE on-chain: escrowed, verified, and royalty-routed, with no human in the loop.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/labor-market. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /labor-market: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/173-swarm-100-route-labor-market.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/174-swarm-100-route-labs.md b/prompts/finish/174-swarm-100-route-labs.mddeleted file mode 100644index 6b6bf927c..000000000--- a/prompts/finish/174-swarm-100-route-labs.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /labs (Labs Showcase)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/labs')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/labs" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/labs--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /labs ("Labs Showcase", section "labs").-Declared purpose: A curated gallery of three.ws's most powerful: and most hidden: features. Find and try things you didn't know existed.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/labs. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /labs: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/174-swarm-100-route-labs.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/175-swarm-100-route-launcher.md b/prompts/finish/175-swarm-100-route-launcher.mddeleted file mode 100644index d0b3c4938..000000000--- a/prompts/finish/175-swarm-100-route-launcher.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /launcher (Memetic Launcher)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/launcher')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/launcher" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/launcher--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /launcher ("Memetic Launcher", section "crypto").-Declared purpose: Your personal autonomous coin launcher: set your mode (trend, meme, hybrid, random), pick trend sources, tune your cadence, preview what your agents would mint, then go live: real pump.fun launches on your schedule, self-funded from your own agents' wallets under a hard daily SOL cap.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/launcher. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /launcher: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/175-swarm-100-route-launcher.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/176-swarm-100-route-launchpad.md b/prompts/finish/176-swarm-100-route-launchpad.mddeleted file mode 100644index e46329085..000000000--- a/prompts/finish/176-swarm-100-route-launchpad.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /launchpad (Launchpad Studio)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/launchpad')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/launchpad" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/launchpad--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /launchpad ("Launchpad Studio", section "labs").-Declared purpose: Build a hosted 3D launchpad, token page, concierge, or showroom on a three.ws subdomain.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/launchpad. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /launchpad: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/176-swarm-100-route-launchpad.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/177-swarm-100-route-leaderboard.md b/prompts/finish/177-swarm-100-route-leaderboard.mddeleted file mode 100644index 858638cbe..000000000--- a/prompts/finish/177-swarm-100-route-leaderboard.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /leaderboard (Trader Leaderboard)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/leaderboard')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/leaderboard" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/leaderboard--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /leaderboard ("Trader Leaderboard", section "crypto").-Declared purpose: Pump.fun traders ranked by a provable, on-chain track record: realized P&L, win rate, drawdown: with every number traceable to its transaction.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/leaderboard. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /leaderboard: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/177-swarm-100-route-leaderboard.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/178-swarm-100-route-ledger.md b/prompts/finish/178-swarm-100-route-ledger.mddeleted file mode 100644index 0a7858693..000000000--- a/prompts/finish/178-swarm-100-route-ledger.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /ledger (Reasoning Ledger)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/ledger')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/ledger" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/ledger--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /ledger ("Reasoning Ledger", section "agent-tools").-Declared purpose: An agent's every consequential decision (what it decided, why, what it predicted, and what actually happened) in a tamper-evident, on-chain-anchored, independently verifiable ledger. Open any agent's at /ledger/<agentId>. Being wrong is visible; honesty is the trust signal.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/ledger. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /ledger: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/178-swarm-100-route-ledger.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/179-swarm-100-route-lipsync-mic.md b/prompts/finish/179-swarm-100-route-lipsync-mic.mddeleted file mode 100644index c1ca0717a..000000000--- a/prompts/finish/179-swarm-100-route-lipsync-mic.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /lipsync/mic (Lipsync (Mic))--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/lipsync/mic')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/lipsync/mic" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/lipsync/mic--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /lipsync/mic ("Lipsync (Mic)", section "labs").-Declared purpose: Feed live mic audio through the viseme analyser and watch the avatar's mouth track your voice in real time.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/lipsync/mic. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /lipsync/mic: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/179-swarm-100-route-lipsync-mic.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/180-swarm-100-route-lipsync.md b/prompts/finish/180-swarm-100-route-lipsync.mddeleted file mode 100644index 0888463d5..000000000--- a/prompts/finish/180-swarm-100-route-lipsync.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /lipsync (Lipsync (TTS))--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/lipsync')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/lipsync" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/lipsync--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /lipsync ("Lipsync (TTS)", section "labs").-Declared purpose: Type text, stream it through the avatar's voice, and the mouth animates in real time via wawa-lipsync viseme detection.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/lipsync. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /lipsync: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/180-swarm-100-route-lipsync.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/181-swarm-100-route-live.md b/prompts/finish/181-swarm-100-route-live.mddeleted file mode 100644index 2971cf7b2..000000000--- a/prompts/finish/181-swarm-100-route-live.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /live (Agent Economy: Live)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/live')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/live" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/live--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /live ("Agent Economy: Live", section "crypto").-Declared purpose: Two AI agents in a virtual world paying each other for live Solana market data. Real on-chain transactions, real blockchain, real 3D.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/live. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /live: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/181-swarm-100-route-live.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/182-swarm-100-route-lookup.md b/prompts/finish/182-swarm-100-route-lookup.mddeleted file mode 100644index 601329dd3..000000000--- a/prompts/finish/182-swarm-100-route-lookup.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /lookup (Agent Lookup)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/lookup')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/lookup" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/lookup--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /lookup ("Agent Lookup", section "crypto").-Declared purpose: Resolve any three.ws agent by Solana mint, agent ID, avatar ID, or slug: renders its interactive 3D avatar alongside its on-chain identity (collection, owner, agent wallet, Active/x402 status, Metaplex/Solscan/Magic Eden links).--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/lookup. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /lookup: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/182-swarm-100-route-lookup.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/183-swarm-100-route-marketplace-analytics.md b/prompts/finish/183-swarm-100-route-marketplace-analytics.mddeleted file mode 100644index 1ce32489f..000000000--- a/prompts/finish/183-swarm-100-route-marketplace-analytics.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /marketplace/analytics (Marketplace Analytics)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/marketplace/analytics')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/marketplace/analytics" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/marketplace/analytics--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /marketplace/analytics ("Marketplace Analytics", section "main").-Declared purpose: Real-time stats on the three.ws skill marketplace: top skills, top agents, and sales volume.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/marketplace/analytics. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /marketplace/analytics: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/183-swarm-100-route-marketplace-analytics.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/184-swarm-100-route-markets-archive.md b/prompts/finish/184-swarm-100-route-markets-archive.mddeleted file mode 100644index 6eb25e592..000000000--- a/prompts/finish/184-swarm-100-route-markets-archive.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /markets/archive (Crypto News Archive)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/markets/archive')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/markets/archive" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/markets/archive--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /markets/archive ("Crypto News Archive", section "crypto").-Declared purpose: The largest open crypto-news archive: 660,000+ articles from September 2017 to today, refreshed hourly from the live feed, in English and Chinese: searchable by keyword, ticker, source, sentiment, date, and language.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/markets/archive. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /markets/archive: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/184-swarm-100-route-markets-archive.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/185-swarm-100-route-markets-digest.md b/prompts/finish/185-swarm-100-route-markets-digest.mddeleted file mode 100644index 420d71121..000000000--- a/prompts/finish/185-swarm-100-route-markets-digest.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /markets/digest (Crypto News Digest)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/markets/digest')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/markets/digest" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/markets/digest--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /markets/digest ("Crypto News Digest", section "crypto").-Declared purpose: The day's crypto news grouped into the handful of stories that actually matter: each with a plain-language summary, a market stance, the tickers involved, and every outlet that covered it. Clustered from 38 live publisher feeds.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/markets/digest. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /markets/digest: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/185-swarm-100-route-markets-digest.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/186-swarm-100-route-markets-news-article.md b/prompts/finish/186-swarm-100-route-markets-news-article.mddeleted file mode 100644index 356ec6dbd..000000000--- a/prompts/finish/186-swarm-100-route-markets-news-article.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /markets/news/article (Crypto News Reader)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/markets/news/article')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/markets/news/article" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/markets/news/article--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /markets/news/article ("Crypto News Reader", section "crypto").-Declared purpose: Rich article reader for crypto news: full story extraction with AI summary, key points, sentiment, detected tickers, and related coverage from across the industry.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/markets/news/article. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /markets/news/article: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/186-swarm-100-route-markets-news-article.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/187-swarm-100-route-markets-news.md b/prompts/finish/187-swarm-100-route-markets-news.mddeleted file mode 100644index 23ca7acd9..000000000--- a/prompts/finish/187-swarm-100-route-markets-news.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /markets/news (Crypto News)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/markets/news')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/markets/news" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/markets/news--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /markets/news ("Crypto News", section "crypto").-Declared purpose: Live crypto news aggregated natively from 192 publisher feeds in 18 languages: filter by category (Bitcoin, DeFi, security, regulation, and more), pick a language, search headlines, and open any story in a rich reader with summary, key points, sentiment, and related coverage.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/markets/news. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /markets/news: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/187-swarm-100-route-markets-news.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/188-swarm-100-route-markets-robinhood.md b/prompts/finish/188-swarm-100-route-markets-robinhood.mddeleted file mode 100644index 990276899..000000000--- a/prompts/finish/188-swarm-100-route-markets-robinhood.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /markets/robinhood (Robinhood Chain)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/markets/robinhood')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/markets/robinhood" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/markets/robinhood--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /markets/robinhood ("Robinhood Chain", section "crypto").-Declared purpose: The 24/7 tokenized-equity board for Robinhood Chain (4663): live Chainlink NAV vs DEX premium for every Stock Token, a memecoin screener across NOXA and The Odyssey launchpads, recent launches, and chain TVL: plus a real wallet-connect buy flow for its coins.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/markets/robinhood. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /markets/robinhood: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/188-swarm-100-route-markets-robinhood.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/189-swarm-100-route-markets-trending.md b/prompts/finish/189-swarm-100-route-markets-trending.mddeleted file mode 100644index 3857852c8..000000000--- a/prompts/finish/189-swarm-100-route-markets-trending.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /markets/trending (Trending Crypto)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/markets/trending')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/markets/trending" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/markets/trending--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /markets/trending ("Trending Crypto", section "crypto").-Declared purpose: The most-searched coins, categories, and NFTs on CoinGecko over the last 24 hours: trending rank, price, 24h change, market cap, and volume, each coin and category linking to its detail page. Auto-refreshing.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/markets/trending. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /markets/trending: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/189-swarm-100-route-markets-trending.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/190-swarm-100-route-markets.md b/prompts/finish/190-swarm-100-route-markets.mddeleted file mode 100644index bf840a8ce..000000000--- a/prompts/finish/190-swarm-100-route-markets.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /markets (Markets Hub)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/markets')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/markets" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/markets--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /markets ("Markets Hub", section "crypto").-Declared purpose: The three.ws markets hub: live global stats, the top 100 coins, breaking crypto news from 192 publisher feeds, a 662k-article news archive, and hero links to the full markets toolset.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/markets. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /markets: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/190-swarm-100-route-markets.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/191-swarm-100-route-meta-allocator.md b/prompts/finish/191-swarm-100-route-meta-allocator.mddeleted file mode 100644index 8c5c3fa3b..000000000--- a/prompts/finish/191-swarm-100-route-meta-allocator.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /meta-allocator (Meta-Allocator: the ETF of degens)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/meta-allocator')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/meta-allocator" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/meta-allocator--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /meta-allocator ("Meta-Allocator: the ETF of degens", section "crypto").-Declared purpose: Do not pick a coin, pick a basket of proven traders. The Meta-Allocator spreads a budget across the top verified pump.fun trader agents, ranked by risk-adjusted on-chain track record and diversified across trading styles, then tells you how to rebalance. Non-custodial: it returns a plan you act on one leader at a time, never a transaction.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/meta-allocator. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /meta-allocator: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/191-swarm-100-route-meta-allocator.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/192-swarm-100-route-minted.md b/prompts/finish/192-swarm-100-route-minted.mddeleted file mode 100644index d0f8aa295..000000000--- a/prompts/finish/192-swarm-100-route-minted.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /minted (Minted 3D Assets)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/minted')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/minted" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/minted--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /minted ("Minted 3D Assets", section "crypto").-Declared purpose: Live public gallery of every generated 3D avatar minted as a Metaplex Core NFT through three.ws: interactive viewer, baked provenance, and enforced creator royalties on every card.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/minted. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /minted: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/192-swarm-100-route-minted.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/193-swarm-100-route-mirror.md b/prompts/finish/193-swarm-100-route-mirror.mddeleted file mode 100644index 5b7484ca4..000000000--- a/prompts/finish/193-swarm-100-route-mirror.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /mirror (Copy Trading)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/mirror')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/mirror" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/mirror--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /mirror ("Copy Trading", section "crypto").-Declared purpose: Follow a high-performing agent and your own agent mirrors its real trades, proportionally and strictly within your spend policy.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/mirror. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /mirror: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/193-swarm-100-route-mirror.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/194-swarm-100-route-mocap-studio.md b/prompts/finish/194-swarm-100-route-mocap-studio.mddeleted file mode 100644index 647066987..000000000--- a/prompts/finish/194-swarm-100-route-mocap-studio.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /mocap-studio (Mocap Studio)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/mocap-studio')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/mocap-studio" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/mocap-studio--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /mocap-studio ("Mocap Studio", section "main").-Declared purpose: Drive a 3D avatar with your webcam: real-time facial motion capture in the browser, no rig and no download.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/mocap-studio. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /mocap-studio: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/194-swarm-100-route-mocap-studio.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/195-swarm-100-route-monitor.md b/prompts/finish/195-swarm-100-route-monitor.mddeleted file mode 100644index 808f54c37..000000000--- a/prompts/finish/195-swarm-100-route-monitor.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /monitor (Agent Monitor)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/monitor')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/monitor" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/monitor--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /monitor ("Agent Monitor", section "build").-Declared purpose: Ops-room dashboard for the 3D AI agent fleet: live agent activity with a 3D spotlight, on-chain money pulse, x402 revenue, agent-to-agent hires, coin launches, the community forge feed, crews and platform health on one screen.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/monitor. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /monitor: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/195-swarm-100-route-monitor.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/196-swarm-100-route-motion-swap.md b/prompts/finish/196-swarm-100-route-motion-swap.mddeleted file mode 100644index 9b829f954..000000000--- a/prompts/finish/196-swarm-100-route-motion-swap.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /motion-swap (Motion Swap: Replace Yourself in Video with Your Avatar)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/motion-swap')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/motion-swap" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/motion-swap--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /motion-swap ("Motion Swap: Replace Yourself in Video with Your Avatar", section "build").-Declared purpose: Upload a video of yourself and get it back with your 3D avatar performing your motion instead of you. Server-side pose tracking and person masking, browser-side compositing and export: stay off camera without staying off video.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/motion-swap. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /motion-swap: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/196-swarm-100-route-motion-swap.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/197-swarm-100-route-nim-forge.md b/prompts/finish/197-swarm-100-route-nim-forge.mddeleted file mode 100644index 4d6197553..000000000--- a/prompts/finish/197-swarm-100-route-nim-forge.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /nim-forge (TRELLIS NIM Forge)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/nim-forge')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/nim-forge" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/nim-forge--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /nim-forge ("TRELLIS NIM Forge", section "build").-Declared purpose: Drop in a photo, get a textured 3D model, reconstructed on a self-hosted Microsoft TRELLIS NIM. Single-hop image to GLB.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/nim-forge. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /nim-forge: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/197-swarm-100-route-nim-forge.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/198-swarm-100-route-notifications.md b/prompts/finish/198-swarm-100-route-notifications.mddeleted file mode 100644index 6e0475f14..000000000--- a/prompts/finish/198-swarm-100-route-notifications.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /notifications (Notifications)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/notifications')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/notifications" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/notifications--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /notifications ("Notifications", section "main").-Declared purpose: Every payment, sale, remix, follow, quest, and royalty notification on three.ws: all in one place, filterable by category.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/notifications. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /notifications: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/198-swarm-100-route-notifications.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/prompts/finish/199-swarm-100-route-nvidia.md b/prompts/finish/199-swarm-100-route-nvidia.mddeleted file mode 100644index 01c773eee..000000000--- a/prompts/finish/199-swarm-100-route-nvidia.md+++ /dev/null@@ -1,119 +0,0 @@-# Route audit: /nvidia (three.ws on NVIDIA)--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If sibling swarm-100 files in-prompts/finish/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. This route's declared entry (path, description, added date):-- node -e "const p=require('./data/pages.json');for(const s of p.sections)for(const g of s.pages)if(g.path==='/nvidia')console.log(JSON.stringify(g,null,2))"--2. Find the serving implementation: grep "/nvidia" in vercel.json (the live- route table server/index.mjs boots from), then list matching files in- pages/ and src/. STRUCTURE.md maps every surface to its directory if the- grep is ambiguous.-3. Production right now:-- curl -so /dev/null -w '%{http_code}\n' https://three.ws/nvidia--4. Start (or reuse) the local dev server: npm run dev (vite, port 3000).--## Task: audit this one route end to end, then fix everything found--Route: /nvidia ("three.ws on NVIDIA", section "main").-Declared purpose: three.ws is a member of the NVIDIA Inception program. Every 3D model, avatar, and voice on the platform runs on NVIDIA: a free hosted lane behind one API key covering text-to-3D, chat, vision, embeddings, safety and speech, plus a self-hosted fleet of Cloud Run GPU workers on L4 and RTX PRO 6000.--Drive the page in a real headless browser. Playwright is already a dev-dependency; keep your probe script in the session scratchpad, never in the-repo.--1. Load http://localhost:3000/nvidia. Record the HTTP status, every console- error and warning, and every failed network request.-2. Exercise it like a first-time user: click the primary actions, run the- primary flow if one exists, follow the page's internal links. Every button- must do something real; every link must resolve. If the page renders data,- prove in the network log that it came from a live endpoint, not a baked-in- fallback array.-3. States: verify populated, empty, loading, and error states are designed and- reachable. Block the page's primary fetch in Playwright and confirm an- actionable error state appears, not a blank void or a raw exception.-4. Responsive: render at 320, 768, and 1440 px. No horizontal page scroll, no- overlapping or clipped controls.-5. Accessibility basics: one h1; interactive elements are real buttons or- links with visible focus states; images carry alt text; form fields have- labels.-6. Head: a specific title tag and meta description that match the declared- purpose above.-7. Fix every defect at its root in the page's real source files and re-run the- probe until clean. Defects in shared modules are in scope when this route- exercises them.--## Definition of done--- [ ] Local load of /nvidia: HTTP 200, zero console errors, zero unhandled- failed requests (a third-party outage must land in a designed fallback,- not a broken page).-- [ ] No dead interactive element on the page: no stub href="#", no button- without a handler, no link to a missing path.-- [ ] Empty, loading, error, and populated states verified.-- [ ] 320 / 768 / 1440 px renders verified clean.-- [ ] Title and meta description present and specific.-- [ ] npm test passes; npm run check:rules -- --paths <touched files> exits 0.-- [ ] Production status from step 0 recorded in the report; if production- needs a deploy to pick up a fix, the report says so (deploys stay- owner-gated).--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/finish/199-swarm-100-route-nvidia.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.