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
docs(prompts): retire the forge UX work order now its generation flow has landed
nirholasa149af75b72e
diff --git a/prompts/RUN-ORDER.md b/prompts/RUN-ORDER.mdindex 36dea3276..19d8635a2 100644--- a/prompts/RUN-ORDER.md+++ b/prompts/RUN-ORDER.md@@ -73,7 +73,6 @@ Run in this order; each is one session. This tier is what a visitor actually jud|---|---|| [finish/quality-bar-07-design-system-sweep.md](finish/quality-bar-07-design-system-sweep.md) | Actively regressing: raw hex went 4,787 to 5,426 since 2026-08-02 and `audit:tokens` reports 10 against a baseline of 0. Every day this waits, more surfaces are built off-token. || [finish/quality-bar-04-pbr-texture-material-realism.md](finish/quality-bar-04-pbr-texture-material-realism.md) | `api/_lib/glb-pbr-derive.js` is imported and never called, so every generated model ships without derived PBR. This is the single biggest quality delta available on the core product. |-| [finish/quality-bar-06-forge-ux-flow.md](finish/quality-bar-06-forge-ux-flow.md) | The forge is the front door. The result-moment click-through table and the audits are still owed. || [finish/quality-bar-08-mobile-performance.md](finish/quality-bar-08-mobile-performance.md) | Touch targets shipped; the after-table, default GLB compression and the WebGL budget are open. Mobile is where 3D fails first. || [finish/swarm-100-sweep-perf.md](finish/swarm-100-sweep-perf.md) | Eight pages measured 2026-08-15, three under 80, no re-measure since the 2026-09-01 fixes. Cheap once P0 ships. || [finish/quality-bar-03-gpu-fleet-scaleout.md](finish/quality-bar-03-gpu-fleet-scaleout.md) | Cold-start UX shipped; the load test and scale ceilings are open. Credits are pre-approved, so nothing here is gated. |diff --git a/prompts/finish/004-quality-bar-06-forge-ux-flow.md b/prompts/finish/004-quality-bar-06-forge-ux-flow.mddeleted file mode 100644index f881e9223..000000000--- a/prompts/finish/004-quality-bar-06-forge-ux-flow.md+++ /dev/null@@ -1,102 +0,0 @@-# QB-06: Forge UX, from prompt to proud screenshot--**How to run this:** paste this whole file into a fresh Claude Code chat opened in-`/workspaces/three.ws`, or say "execute `prompts/finish/004-quality-bar-06-forge-ux-flow.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. Blockers have pre-answered routes at the bottom. Use them and keep going.-3. CLAUDE.md hard rules: no mocks, no stubs, no TODO comments, no fake progress bars, no- em-dash or en-dash characters. Stage explicit paths only. Do not push unless asked.--## Mission--Make `/forge` feel like a flagship product: the path from typing a prompt to holding a-share-worthy 3D result should be legible, alive, and impossible to dead-end.--## Step 0: re-derive current state (trust nothing below)--```bash-npm run dev # port 3000, then walk /forge as a first-time user in a real browser-ls src/forge-compare.js src/forge-prompt-gen.js api/_lib/forge-lane-health.js-grep -n "starter\|chip" pages/forge.html | head -20-npm run audit:web -- /forge # page-audit takes explicit routes as positional args-```--Known shipped as of 2026-07-30, verify then skip: generation history persistence, the-side-by-side compare mode (`src/forge-compare.js`, `tests/forge-compare.test.js`, documented in-`docs/forge.md`), and the six curated starter prompts on `pages/forge.html`.--## Tasks--1. **Walk the flow as a first-time user** (`npm run dev`, real browser, also at 320 px). Write- down every moment of confusion: what is happening now, how long will it take, what do I do- with the result. Fix what you find. The list below is the floor, not the ceiling.-2. **Generation timeline.** Replace passive waiting with a real stage timeline (enhancing- prompt, generating reference views, sculpting mesh, painting textures, finishing) driven by- the poll states the API already returns, with elapsed time and the honest ETA from- `api/_lib/forge-lane-health.js`. Show reference images as they are produced; watching the- concept appear kills perceived latency.-3. **Result moment.** When the GLB lands: a camera intro move, the quality viewer, and- immediate actions with zero hunting: download GLB and USDZ, view in AR (`/ar` handoff), rig- it, refine it (`refine_model`), restyle materials, place it IRL (`/irl`), share a link whose- og-image renders this model. Click every one of those through to its destination and back.- A dead button is a failed task.-4. **Prompt help.** After generation, show the director's enhanced prompt ("what we actually- asked the model") with a one-click "try a variation".-5. **Error and retry UX.** Every failure path renders a designed state: what failed in plain- language, the automatic recovery that already happened (failover), and the one-click manual- options (engine switch). No raw error strings, no dead ends. Session expiry mid-generation- must preserve the job and resume after re-auth; test that by clearing the session cookie- mid-poll.-6. **Keyboard and accessibility.** Cmd/Ctrl+Enter submits, Esc collapses panels, focus-visible- everywhere, ARIA labels on viewer controls, `prefers-reduced-motion` respected on the intro- move. Confirm with `npm run audit:a11y`.--## Definition of done--- [ ] The full walk-through repeats clean: no confusion points, no console errors, loading,- empty, error and success states designed at 320, 768 and 1440 px.-- [ ] Every result-moment action clicked through to its destination feature and back, listed- one by one in the report with what happened.-- [ ] `npm run audit:web` clean on `/forge`; `npm run audit:a11y` no new failures.-- [ ] `npm test` green (do not pipe through `tail`).-- [ ] `data/changelog.json` entry in holder language; screenshots of the result moment in the- report.-- [ ] `npm run check:rules -- --paths <files you touched>` clean.--## Never blocked (pre-answered)--| Blocker | Do this |-|---|---|-| A poll payload lacks a stage you want to show | Extend the response in `api/forge.js` as an additive field (old clients keep working). Never fake a state client-side. |-| USDZ missing for a model | Generate on demand through the existing bake pipeline and show a brief honest "preparing AR file" state. |-| Persistence store choice | Use what `api/` already uses for user artifacts (R2 plus database rows). Do not introduce new storage. |-| The crawler reports WebGL or texture errors | `npm run audit:web` runs pages concurrently and produces false 3D failures. Re-check any 3D failure serially before reporting it. |-| A generation is slow because a GPU is cold | That is task 2's subject, not a blocker. Show the honest ETA. |--## Report format--The confusion list with what you changed for each, the action-by-action click-through table,-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/004-quality-bar-06-forge-ux-flow.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/_context/production-100-00-INDEX.md b/prompts/finish/_context/production-100-00-INDEX.mdindex 69ebf614b..ca6f667fd 100644--- a/prompts/finish/_context/production-100-00-INDEX.md+++ b/prompts/finish/_context/production-100-00-INDEX.md@@ -83,7 +83,6 @@ run in parallel, H waits on owner gates, I is anytime, J is the parallel swarm.|---|---|| [../quality-bar/03-gpu-fleet-scaleout.md](../006-quality-bar-03-gpu-fleet-scaleout.md) | Partial: cold-start UX and keep-warm shipped; load test and scale ceilings open. || [../quality-bar/04-pbr-texture-material-realism.md](../003-quality-bar-04-pbr-texture-material-realism.md) | Mostly open; `api/_lib/glb-pbr-derive.js` exists and is imported but never called. |-| [../quality-bar/06-forge-ux-flow.md](../004-quality-bar-06-forge-ux-flow.md) | Partial: the result-moment click-through table and the audits are still owed. || [../quality-bar/07-design-system-sweep.md](../002-quality-bar-07-design-system-sweep.md) | Open and regressed: raw-hex count 4,787 to 5,426 since 2026-08-02; `audit:tokens` 10 vs a baseline of 0. || [../quality-bar/08-mobile-performance.md](../005-quality-bar-08-mobile-performance.md) | Partial: touch targets shipped; the after-table, default GLB compression and the WebGL budget are open. || [../quality-bar/10-avatar-likeness-irl-people.md](../007-quality-bar-10-avatar-likeness-irl-people.md) | Partial: animation dignity 10/10 rigs; the likeness audit's only complete run shows no improvement. |diff --git a/prompts/finish/_context/quality-bar-README.md b/prompts/finish/_context/quality-bar-README.mdindex b4919ac13..aa8fcdabe 100644--- a/prompts/finish/_context/quality-bar-README.md+++ b/prompts/finish/_context/quality-bar-README.md@@ -20,17 +20,19 @@ special case.|---|------|------|-------|| 03 | [03-gpu-fleet-scaleout.md](../006-quality-bar-03-gpu-fleet-scaleout.md) | Scale ceilings, cold-start honesty, keep-warm, load test | Partial. Cold-start honesty (task 2) and keep-warm (task 3) are DONE and verified; do not redo them. Scale ceilings (task 1) need one `gcloud auth login`. The load test (tasks 4-5) is blocked by a production outage, not by capacity: every `/api/forge` generation fails on a rejected object-storage credential, so nothing can be measured until it is rotated and the pending commits deploy. || 04 | [04-pbr-texture-material-realism.md](../003-quality-bar-04-pbr-texture-material-realism.md) | Full PBR sets, measured presets, skin/eye/hair | Partial. Everything but the bench delta is verified: the channel matrix, the derive pass wired opt-out into `forge-store.js`, the presets rendered under all four shipped HDRIs (`docs/assets/material-presets/`), the avatar skin/eye/hair readback at 320/768/1440 (the check had been aimed at `/viewer`, which never runs that pass, and had been reporting zero meshes), and the mobile payload gate with measured 4K byte sizes (`npm run check:glb:payload`). The quality-bench comparison needs Vertex, and this Codespace's gcloud credentials have expired: `gcloud auth application-default login` unblocks it. |-| 06 | [06-forge-ux-flow.md](../004-quality-bar-06-forge-ux-flow.md) | The `/forge` flagship experience | Partial. History, compare mode and starters shipped; the result-moment click-through and the clean audit sweep are open. || 07 | [07-design-system-sweep.md](../002-quality-bar-07-design-system-sweep.md) | Tokens, states, microinteractions sitewide | Open || 08 | [08-mobile-performance.md](../005-quality-bar-08-mobile-performance.md) | Mobile excellence, GLB delivery | Partial. A measured baseline exists in `_generated/08/`; the fixes and the re-measure are open. || 10 | [10-avatar-likeness-irl-people.md](../007-quality-bar-10-avatar-likeness-irl-people.md) | IRL people: likeness, hands, rig, AR | Partial. Animation dignity is proven across ten rig conventions; one runtime-lane defect plus the likeness audit are open. |Retired after verification (readable in git history): 01 photoreal reference pipeline, 02 the-Hunyuan3D flagship lane, 05 cinematic viewers, 09 the realism eval harness (now-`scripts/quality-bench.mjs` plus the weekly cron).+Hunyuan3D flagship lane, 05 cinematic viewers, 06 the `/forge` generation flow (the pre-submit+progress channel, `api/_lib/forge-progress.js`, so the reference view appears while the POST is+still open; the stacked-layout reveal; the designed storage-outage state; reduced-motion and the+result-moment click-through), 09 the realism eval harness (now `scripts/quality-bench.mjs` plus+the weekly cron).Suggested order: 04, 07 and 08 have no dependencies and can run in parallel. 03 unblocks-throughput. 06 and 10 land best after 04.+throughput. 10 lands best after 04.## Ground rules recap (full versions in `quality-bar-_shared.md` and `CLAUDE.md`)