Skip to content

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.

docs(prompts): move the OpenAI Apps SDK handoff pack under prompts/ and retire its five shipped briefs

nirholas09bfbb1b5d08
diff --git a/docs/big-tech-recognition-week.md b/docs/big-tech-recognition-week.md
index 640f6e5b8..53a31e855 100644
--- a/docs/big-tech-recognition-week.md
+++ b/docs/big-tech-recognition-week.md
@@ -76,11 +76,11 @@ submission target is the free keyless connector at
Actions surface, all of which are live and re-verified.
- **Package:** [`prompts/store-submissions/_generated/openai-submission.md`](../prompts/store-submissions/_generated/openai-submission.md) (7/7 policy PASS, real screenshots, real captured tool responses)
-- **Pre-submission verification pack:** [`docs/openai-pr/00-START-HERE.md`](./openai-pr/00-START-HERE.md), with task 07 as the go/no-go gate
+- **Pre-submission verification pack:** [`prompts/openai-pr/00-START-HERE.md`](../prompts/openai-pr/00-START-HERE.md), with task 07 as the go/no-go gate
- **Standing status:** Partner Network accepted 2026-07-14; organization verified; GPT Store listing live and public
- **Open item:** the final submit in the partner portal
-Before the submit, run [`docs/openai-pr/07-final-verification-and-submit.md`](./openai-pr/07-final-verification-and-submit.md) against the currently deployed
+Before the submit, run [`prompts/openai-pr/07-final-verification-and-submit.md`](../prompts/openai-pr/07-final-verification-and-submit.md) against the currently deployed
revision. Its checklist exists precisely because the answer sheet asserts
"verified live" facts dated 2026-07-14 and deployments move. That is a
one-session job, not a blocker.
@@ -285,7 +285,7 @@ Ordered so that the items with the longest external latency go out first.
| Day | Action | Owner |
|---|---|---|
-| Mon 08-17 | Run [openai-pr task 07](./openai-pr/07-final-verification-and-submit.md) against the live revision, then **submit the OpenAI App Directory package** | Owner (portal) |
+| Mon 08-17 | Run [openai-pr task 07](../prompts/openai-pr/07-final-verification-and-submit.md) against the live revision, then **submit the OpenAI App Directory package** | Owner (portal) |
| Mon 08-17 | Approve and run the staged **MCP Registry** publish; submit **`three-ws-3d`** to the Claude plugin marketplace | Owner (approval), agent (commands) |
| Tue 08-18 | Correct the **NVIDIA Inception portal product record** | Owner |
| Tue 08-18 | Send the **five-ask Inception email** to `[email protected]` | Owner |
@@ -328,4 +328,4 @@ Publishing, posting, and sending are owner-gated under stop-and-ask gate 2 in
- [Listings and distribution](./listings.md): the canonical program and directory inventory
- [Announcement coverage](./announcement-coverage.md): which surfaces have been announced on X and which have not
- [Press kit](./press-kit.md): marks, boilerplate, and the rules governing co-branded graphics
-- [OpenAI submission handoff pack](./openai-pr/00-START-HERE.md): the pre-submission task briefs
+- [OpenAI submission handoff pack](../prompts/openai-pr/00-START-HERE.md): the pre-submission task briefs
diff --git a/docs/openai-pr/00-START-HERE.md b/docs/openai-pr/00-START-HERE.md
deleted file mode 100644
index 4cf91a0da..000000000
--- a/docs/openai-pr/00-START-HERE.md
+++ /dev/null
@@ -1,90 +0,0 @@
-# OpenAI Apps SDK submission — agent handoff pack
-
-This directory contains standalone task briefs. Each `NN-*.md` file is a complete
-prompt you can hand to one agent. They are ordered by impact and are mostly
-independent, so several can run in parallel. Read this file first, then the brief
-for your task.
-
-## What we are shipping
-
-three.ws is submitting its **3D Studio** app to OpenAI (ChatGPT Apps SDK / App
-Directory). The submission target is the **free, keyless MCP connector** at
-`https://three.ws/api/mcp-studio` plus its inline widget and the custom-GPT
-Actions surface. A reviewer at OpenAI will exercise these surfaces and read the
-manifests we expose. Everything they can reach must be correct, consistent, and
-free of any paid or crypto surface.
-
-The submission kit already exists at
-[`prompts/store-submissions/_generated/`](../../prompts/store-submissions/_generated/)
-(answer sheet `openai-submission.md`, `openai-actions.yaml`, `TRACKER.md`,
-screenshots, evidence JSON). Your job is not to recreate it. Your job is to close
-the specific gaps that would fail review or embarrass us in front of OpenAI.
-
-## The map of the surface (so you know what you are touching)
-
-| Surface | Where it lives | Notes |
-| --- | --- | --- |
-| Free MCP connector (submission target) | `api/mcp-studio.js`, `api/_mcp-studio/*` | Keyless. 10 tools (6 generation + `check_job` + 3 persona). Real forge lane. |
-| Inline ChatGPT widget (the real one) | `api/_mcp-studio/component.js` | `<model-viewer>` skybridge resource `ui://widget/three-studio-model.html` |
-| ~~Orphaned three.js widget~~ | Deleted | Was `apps-sdk/studio-viewer/*` + `scripts/build-apps-sdk-viewer.mjs` + `public/apps-sdk/*`, wired to nothing. Removed 2026-07-18 (task 01, Option A). Do not recreate it. |
-| Embodiment embed | `apps-sdk/embodiment/*`, `pages/embodiment/embed.html` | Persona widget iframes this. |
-| AR-in-ChatGPT | `api/ar.js`, `api/_lib/ar-launch.js` | Every generation carries `arUrl`. |
-| Custom GPT Actions (REST) | `api/3d/studio.js`, `prompts/store-submissions/_generated/openai-actions.yaml` | GPT Store listing is live. |
-| Served manifests | `public/.well-known/*` | Publicly discoverable. See task 02. |
-| Paid MCP server (NOT the submission) | `api/mcp-3d.js`, `api/_mcp3d/*` | x402/OAuth gated. Out of scope unless a task says so. |
-
-## The tasks
-
-| # | Brief | Priority | Blocks submit? |
-| --- | --- | --- | --- |
-| 01 | [Resolve the orphaned studio-viewer widget](01-orphan-widget-resolution.md) | P0 | Done 2026-07-18 (Option A: deleted) |
-| 02 | [Reconcile the served `.well-known` manifests with the zero-payment claim](02-wellknown-manifest-conflict.md) | P0 | Yes (reviewer-discoverable crypto surface) |
-| 03 | [Make the `forge_free` tier story true everywhere](03-forge-free-tier-truth.md) | P1 | Yes (submission doc contradicts code) |
-| 04 | [Close the test gaps on the ChatGPT-facing surface](04-openai-surface-test-coverage.md) | P1 | No, but required for "done" |
-| 05 | [Serve the custom-GPT OpenAPI and align legal URLs](05-custom-gpt-openapi-serving.md) | P1 | Partial |
-| 06 | [Fix doc accuracy and reconcile the tool-count story](06-docs-accuracy-reconciliation.md) | P1 | No |
-| 07 | [Final live verification and submit checklist](07-final-verification-and-submit.md) | P2 (runs last) | Yes (the actual go/no-go) |
-
-Run 02 and 03 first of what remains (01 is already done; they change code and docs the later tasks depend on). 04, 05,
-06 can run in parallel after. 07 runs last, after everything else is merged.
-
-## Rules every task must follow (from `CLAUDE.md`)
-
-These are non-negotiable and OVERRIDE any default behavior:
-
-1. **No mocks, no fake data, no placeholders, no TODOs, no stubs, no commented-out
- code.** If you write it, finish it and wire it. Real APIs only.
-2. **Execute, do not interview.** Pick the most reasonable interpretation, do it,
- verify it, and report. Do not end your turn with "want me to proceed?".
-3. **Every state is designed** (loading/empty/error/populated) for any UI you touch.
-4. **Tests must pass:** run `npm test` (do NOT pipe through `tail` — it masks the
- exit code; a vitest failure gates the Playwright stage).
-5. **Changelog:** any user-visible change gets an entry in
- [`data/changelog.json`](../../data/changelog.json) (date, plain-language title +
- summary, tags from feature/improvement/fix/sdk/infra/docs/security). Run
- `npm run build:pages` to regenerate and validate the feed. Internal-only chores
- do not get an entry.
-6. **Docs ship with the feature.** If you touch a surface and its doc is now wrong,
- fix the doc in the same change. New page/route → `data/pages.json`.
-7. **Commit gate — read this before staging anything.** `$THREE`
- (`FeMbDoX7R1Psc4GEcvJdsbNbZA3bfztcyDCatJVJpump`) is the only coin this platform
- promotes. Before any `git commit` whose diff references ANY crypto project other
- than `$THREE` — in code, docs, manifests, tests, or commit text — STOP and get
- explicit owner approval. Task 02 edits a manifest that currently mentions x402/
- USDC/Solana/Base; edits that keep or restructure those references need the owner's
- yes before commit. Do the work, stage nothing that trips the gate, and flag it.
-8. **Do not push, deploy, or submit to OpenAI without owner approval.** Commit
- locally, prepare everything so the ship is one command, and say what remains.
-9. **Never use the em-dash character.** Use a period, comma, colon, or parentheses.
-10. **Concurrent agents share this worktree.** Stage explicit paths only (never
- `git add -A`). Re-check `git status` and `git diff --staged` right before you
- commit.
-
-## Definition of done for the whole pack
-
-- [ ] Every P0/P1 task's own definition of done is met.
-- [ ] `npm test` is green.
-- [ ] No surface a reviewer can reach advertises a paid or crypto capability.
-- [ ] Every doc and manifest describing the ChatGPT surface matches the running code.
-- [ ] `docs/openai-pr/07-final-verification-and-submit.md` checklist passes end to end.
-- [ ] The only remaining steps are the human `[HUMAN: ...]` items in `TRACKER.md`.
diff --git a/docs/openai-pr/01-orphan-widget-resolution.md b/docs/openai-pr/01-orphan-widget-resolution.md
deleted file mode 100644
index 6c4224ac2..000000000
--- a/docs/openai-pr/01-orphan-widget-resolution.md
+++ /dev/null
@@ -1,121 +0,0 @@
-# Task 01 — Resolve the orphaned studio-viewer widget (P0)
-
-> **Resolved 2026-07-18 via Option A** in commit `19c379125` ("Refactor apps-sdk studio viewer and update OpenAI store submission"): `apps-sdk/studio-viewer/src.js`, `scripts/build-apps-sdk-viewer.mjs`, the `build:apps-sdk-viewer` entry in `package.json`, and both generated artifacts under `public/apps-sdk/` were deleted, and `apps-sdk/README.md` plus `ARCHITECTURE.md` were rewritten to point at the real widget (`api/_mcp-studio/component.js`, resource `ui://widget/three-studio-model.html`). `STRUCTURE.md` never referenced the orphan, so it needed no change. Kept as the record of the finding; the file links below point at the now-deleted paths and are left as plain code references.
-
-Read [`00-START-HERE.md`](00-START-HERE.md) first for the shared rules and the map.
-
-## The problem (verified, with evidence)
-
-There are **two** inline-widget implementations in the tree, and they are not
-connected:
-
-1. **The widget ChatGPT actually renders** is the `<model-viewer>` component built
- inline in [`api/_mcp-studio/component.js`](../../api/_mcp-studio/component.js)
- (`COMPONENT_HTML`). It is served as the skybridge resource
- `ui://widget/three-studio-model.html` and is fully wired and tested.
-
-2. **A second, unrelated three.js viewer** lived in
- `apps-sdk/studio-viewer/src.js` (619 lines,
- full GLTFLoader + OrbitControls + RoomEnvironment). It was built by
- `scripts/build-apps-sdk-viewer.mjs`
- into `public/apps-sdk/studio-viewer.bundle.js` and `.../studio-viewer.html`.
- (Both files deleted in `0b50834d7`.)
-
-The second one is **orphaned**:
-
-- The consumer the build script names,
- `api/_mcp3d/studio-viewer-resource.js`, **does not exist** (confirmed:
- `ls` returns "No such file or directory").
-- Nothing serves `ui://widget/studio-viewer.html`. The only references to that URI
- are the source comment in `apps-sdk/studio-viewer/src.js:7` and the build script
- comment in `scripts/build-apps-sdk-viewer.mjs:8`.
-- The `/viewer` route in `vercel.json` maps to `/viewer.html`, a different page, not
- this bundle.
-- The build is not in `prebuild`/`build`, so the committed bundle can silently go
- stale.
-
-On top of that, [`apps-sdk/README.md`](../../apps-sdk/README.md) **describes the
-orphaned viewer as if it were the live ChatGPT widget** (lines 17 and 59 claim the
-bundle is "Bundled into the `ui://widget/studio-viewer.html` skybridge resource
-served to ChatGPT" and "Read at runtime by
-`api/_mcp3d/studio-viewer-resource.js`"). Both statements are false against the
-current tree. For a handoff to OpenAI, a public README that misdescribes our own
-integration is unacceptable.
-
-## Your job
-
-Pick ONE of these two resolutions, implement it fully, and make the README true.
-Default to Option A unless you find a concrete reason the three.js viewer is
-better for our submission (if so, state it in your report and do Option B).
-
-### Option A (default) — Delete the dead path
-
-The live `<model-viewer>` widget in `component.js` already renders GLBs inline in
-ChatGPT, is tested, and is what the screenshots were taken against. The three.js
-viewer duplicates it with no consumer.
-
-1. Delete `apps-sdk/studio-viewer/` (the whole subdir), `scripts/build-apps-sdk-viewer.mjs`,
- the `build:apps-sdk-viewer` script in `package.json`, and the generated
- `public/apps-sdk/studio-viewer.bundle.js` + `public/apps-sdk/studio-viewer.html`
- (and any `dist/apps-sdk/*` copies). Grep first for every reference and remove
- each one so nothing dangles.
-2. Rewrite [`apps-sdk/README.md`](../../apps-sdk/README.md) so the `studio-viewer`
- section is gone and the doc accurately describes the two things that DO ship from
- `apps-sdk/`: the embodiment engine, and a pointer to where the real ChatGPT widget
- lives (`api/_mcp-studio/component.js`). Do not leave a section describing a
- directory you just deleted.
-3. Update [`STRUCTURE.md`](../../STRUCTURE.md) if it references `apps-sdk/studio-viewer`.
-
-### Option B — Wire the three.js viewer as the real widget
-
-Only if you determine the three.js viewer is materially better (PBR lighting, contact
-shadow, animation mixer) than the CDN `<model-viewer>` and worth being the widget.
-
-1. Create the missing resource module (the correct path, given the free connector
- owns the widget, is under `api/_mcp-studio/`, not `api/_mcp3d/`) that reads the
- built bundle and serves it as the `text/html+skybridge` resource. Register it in
- [`api/_mcp-studio/dispatch.js`](../../api/_mcp-studio/dispatch.js) alongside the
- existing `three-studio-model.html` resource and point the tool
- `openai/outputTemplate` at it.
-2. Add `build:apps-sdk-viewer` to the build pipeline (`prebuild` or `build`) so the
- bundle can never go stale, and add it to the esbuild-bundle guard if applicable.
-3. Add a test that the resource serves non-empty HTML containing the bundle and
- carries the required `openai/widgetCSP`/`widgetDescription` metadata.
-4. Make `apps-sdk/README.md` accurate against the new wiring (correct file path,
- correct `ui://` URI, correct build step).
-
-Whichever option: **the CSP must still allowlist exactly the origins the widget
-fetches from** (three.ws, the R2 storage origin, and the CDN if used). Match what
-`api/_mcp-studio/component.js` `componentCsp()` already does.
-
-## Constraints
-
-- Follow every rule in `00-START-HERE.md`.
-- This is not a crypto surface, so the commit gate does not apply here, but still
- stage explicit paths only.
-- No dead references left behind: after your change, `grep -rn "studio-viewer" .`
- (excluding node_modules) must return only intentional, accurate hits.
-
-## Verification
-
-- `grep -rn "apps-sdk/studio-viewer\|studio-viewer-resource\|widget/studio-viewer" . | grep -v node_modules`
- returns only what your chosen option intends (nothing for Option A; the real
- resource wiring for Option B).
-- `npm test` is green (run the full suite, do not pipe through `tail`).
-- For Option B: build the bundle, load the served resource, and confirm it renders a
- GLB. For Option A: confirm the live `<model-viewer>` widget still renders (exercise
- a `forge_free` call against `/api/mcp-studio` and open the resulting `viewerUrl`).
-- `apps-sdk/README.md` reads correctly to someone with zero context and every file
- path and `ui://` URI in it resolves.
-
-## Definition of done
-
-- [ ] One option fully implemented, no half-state.
-- [ ] `apps-sdk/README.md` is accurate; `STRUCTURE.md` updated if needed.
-- [ ] No dangling references to the deleted/created paths.
-- [ ] `npm test` green.
-- [ ] `data/changelog.json` entry if this changed anything user-visible (Option B
- would; Option A is an internal cleanup and doc fix, which is `docs`-tagged only
- if the README is user-facing — use judgment, default to a short `docs`/`infra`
- entry).
-- [ ] Report states which option you chose and why.
diff --git a/docs/openai-pr/02-wellknown-manifest-conflict.md b/docs/openai-pr/02-wellknown-manifest-conflict.md
deleted file mode 100644
index 53f0e310a..000000000
--- a/docs/openai-pr/02-wellknown-manifest-conflict.md
+++ /dev/null
@@ -1,133 +0,0 @@
-# Task 02 — Reconcile the served `.well-known` manifests with the zero-payment claim (P0)
-
-Read [`00-START-HERE.md`](00-START-HERE.md) first. **This task touches a manifest
-that references crypto/x402. The commit gate in `00-START-HERE.md` rule 7 applies:
-do the work, but do NOT commit anything that references a non-$THREE crypto project
-without explicit owner approval. Flag it in your report instead.**
-
-## The problem (verified, with evidence)
-
-Our OpenAI submission answer sheet
-([`prompts/store-submissions/_generated/openai-submission.md`](../../prompts/store-submissions/_generated/openai-submission.md))
-asserts the app has a **zero payment surface** and no crypto capability. But a public,
-reviewer-discoverable manifest at a well-known path says the opposite.
-
-[`public/.well-known/ai-plugin.json`](../../public/.well-known/ai-plugin.json)
-read, until commit `1dbbe0b6d` (2026-08-06) rewrote it (verbatim):
-
-```
-"description_for_human": "three.ws — 3D avatar/world generation and live crypto intelligence APIs, pay-per-call via x402.",
-"description_for_model": "three.ws exposes a 3D model viewer plus a paid pay-per-call API catalog settled via the x402 protocol (HTTP 402, USDC on Solana and Base): ... The machine-readable catalog with prices and schemas is at https://three.ws/.well-known/x402.json ..."
-```
-
-And [`public/.well-known/openapi.yaml`](../../public/.well-known/openapi.yaml) (71KB)
-describes the general REST platform (Auth/Avatars/Keys/OAuth/MCP) and does **not**
-include the 3D Studio endpoints (`/api/3d/studio`, `/api/ar`, `forge_free`).
-
-An OpenAI reviewer hitting `https://three.ws/.well-known/ai-plugin.json` saw a
-crypto/payments pitch, directly contradicting the submission. This was the single
-most likely thing to sink the review.
-
-**Where it stands now.** The served manifest leads with the free lane: its
-`description_for_human` is "Free, keyless 3D avatar and world generation, plus an
-optional paid API catalog for agents.", and `description_for_model` names the free
-surface first (`POST /api/3d/studio`, the MCP lane at `/api/mcp-studio`, the viewer,
-`/api/ar`, the read-only `/api/crypto` index) and points free-lane callers at
-[`public/.well-known/3d-studio-openapi.yaml`](../../public/.well-known/3d-studio-openapi.yaml),
-before describing the paid `/api/x402/*` catalog as separate and optional. `logo_url`
-is `https://three.ws/pwa-512x512.png` and `legal_info_url` is `/legal/tos`. The
-3D Studio schema is served at that path and byte-pinned to
-`prompts/store-submissions/_generated/openai-actions.yaml` by
-`npm run check:studio-openapi` (`scripts/sync-studio-openapi.mjs` regenerates the copy).
-`tests/wellknown-manifests.test.js` asserts the served schema carries no
-crypto/payment token and that the platform manifest never advertises the free lane
-as paid; `tests/api/3d-studio-openapi.test.js` pins the schema to exactly
-`/api/3d/studio` and `/api/ar`, keyless, with the AR parameters `api/ar.js` reads.
-`openapi.yaml` still describes the general platform only, which is correct: the free
-lane has its own artifact.
-
-## Important scoping decision (this is why the task is delicate)
-
-This manifest is **also a real, live product surface** for the general three.ws
-platform and its x402 catalog. It is not junk to be deleted. The `$THREE` /
-platform rules in `CLAUDE.md` say the x402 economy and launch directories are real
-features, not leaks to be stripped. So the goal is **not** to erase the crypto
-platform. The goal is to make sure the OpenAI-facing app is cleanly separable from
-it and that the manifest a reviewer reads for the 3D Studio app does not conflate
-the two.
-
-Do NOT unilaterally gut the platform manifest. Instead, implement the option below
-that keeps both stories true, and surface the decision in your report.
-
-## Your job
-
-Make the OpenAI-app surface and the paid-platform surface cleanly distinct at the
-manifest layer. Recommended approach (implement this unless you find it wrong):
-
-1. **Keep the existing platform manifest honest but move it off the bare
- `ai-plugin.json` default if that is what a ChatGPT reviewer keys on.** Confirm
- whether ChatGPT Apps SDK review actually reads `/.well-known/ai-plugin.json`
- (legacy ChatGPT-plugins manifest) at all, or whether the App Directory review is
- driven purely by the MCP connector metadata and the submitted OpenAPI. Document
- what you find with a source. This determines how much the manifest even matters.
-
-2. **The 3D Studio app must have its own, self-describing, payment-free discovery
- artifact.** The custom-GPT OpenAPI already exists at
- `prompts/store-submissions/_generated/openai-actions.yaml` but is not served
- (Task 05 owns serving it). Coordinate with Task 05: the served, discoverable
- schema for the 3D Studio app must describe only `/api/3d/studio` and `/api/ar`,
- with `auth: none`, no x402, no prices, no chain references.
-
-3. **Fix the misleading fields on the platform manifest that a 3D-app reviewer would
- see as conflicting**, without deleting the real x402 catalog:
- - `logo_url` was `favicon.ico` (`ai-plugin.json:14`). The submission cites a
- 512x512 owned-IP icon, and the manifest now points at `pwa-512x512.png`;
- `tests/wellknown-manifests.test.js` fails if it ever reverts to the favicon.
- - Make it unambiguous, in both the human and model descriptions, which
- capabilities are free and which are paid, so nothing reads as "the 3D app costs
- crypto." The 3D generation lane IS free and keyless. Say so plainly.
-
-4. **Add a test** asserting that whatever manifest describes the 3D Studio app (the
- served Task-05 schema) contains no payment/crypto tokens. Reuse the regex approach
- already in [`tests/mcp-studio.test.js`](../../tests/mcp-studio.test.js) (it asserts
- the MCP connector is crypto-surface-free) and apply it to the served 3D-app schema.
-
-## What NOT to do
-
-- Do not delete `public/.well-known/x402.json`, the x402 catalog, or the paid
- platform's own manifests. Those are real features.
-- Do not strip existing other-coin references as if they were secrets (that
- enforcement is retired per `CLAUDE.md`). The commit gate is about not *committing*
- new non-$THREE references, not about scrubbing the working tree.
-- Do not silently change what the paid platform advertises to its own users.
-
-## Constraints
-
-- Every rule in `00-START-HERE.md`, especially the commit gate (rule 7) and
- no-push/no-submit (rule 8).
-- If your fix requires editing a file whose diff still references x402/USDC/Solana/
- Base/another chain, prepare the edit, run the tests, and STOP before commit with a
- one-line flag to the owner. Do not commit that diff yourself.
-
-## Verification
-
-- `curl -s https://three.ws/.well-known/ai-plugin.json` (or read the file) and
- confirm the 3D-app story is unambiguous and payment-free where it describes the app.
-- The served 3D Studio schema (Task 05) passes the no-crypto-tokens test.
-- `npm test` green.
-- You can articulate, in one paragraph, exactly what an OpenAI reviewer sees at each
- well-known path and why none of it contradicts the submission.
-
-## Definition of done
-
-- [x] The 3D Studio app has a served, payment-free, self-describing discovery artifact
- (jointly with Task 05): `/.well-known/3d-studio-openapi.yaml`.
-- [x] The platform `ai-plugin.json` no longer reads as "the 3D app is a paid crypto
- service"; free vs paid is explicit; logo is a real branded asset.
-- [x] A test asserts the 3D-app schema is crypto/payment-free
- (`tests/wellknown-manifests.test.js`, `tests/api/3d-studio-openapi.test.js`).
-- [ ] `npm test` green.
-- [ ] Any commit that would trip the commit gate is prepared but NOT committed;
- flagged to the owner in your report with the exact diff.
-- [ ] Report documents whether ChatGPT App Directory review actually reads
- `/.well-known/ai-plugin.json`, with a source.
diff --git a/docs/openai-pr/03-forge-free-tier-truth.md b/docs/openai-pr/03-forge-free-tier-truth.md
deleted file mode 100644
index eee30d879..000000000
--- a/docs/openai-pr/03-forge-free-tier-truth.md
+++ /dev/null
@@ -1,121 +0,0 @@
-# Task 03 — Make the `forge_free` tier story true everywhere (P1, blocks submit)
-
-Read [`00-START-HERE.md`](00-START-HERE.md) first.
-
-## The problem (verified, with evidence)
-
-The submission answer sheet and the code disagree about what quality tier
-`forge_free` returns by default.
-
-**The code** ([`api/_mcp-studio/tools.js`](../../api/_mcp-studio/tools.js),
-`handleForgeFree`) defaults to `standard` and degrades an explicit `high` to
-`standard` on a 402 or submit timeout. When this task was written the comment above
-that line said the high tier was blocked on an async worker; today it reads
-(verbatim):
-
-```
-// Standard by default, and every doc describing this tool says exactly that.
-// The high tier is a real, working option, not a stub: it runs on our own
-// async Hunyuan3D worker (GCP_HUNYUAN3D_URL) behind a genuine poll handle,
-// and a live production probe on 2026-08-06 returned a 2.69 MB high-tier GLB
-// from backend `hunyuan3d` end to end. It stays opt-in rather than default
-...
-const tier = VALID_TIER.has(args.tier) ? args.tier : 'standard';
-```
-
-The two reasons it stays opt-in are in that comment: the worker is scale-to-zero,
-so a cold container adds a spin-up on top of the generation, and the high-tier
-access gate is cleared only by the platform seed token (`CRON_SECRET`), so a
-deployment without it would quietly serve standard under a "high by default" promise.
-A job that outlives `STUDIO_FORGE_TIMEOUT_MS` comes back as a pollable handle, never
-an error.
-
-**The submission doc**
-([`prompts/store-submissions/_generated/openai-submission.md`](../../prompts/store-submissions/_generated/openai-submission.md))
-was internally inconsistent:
-- Line ~31 correctly said the tools "default to the standard tier."
-- Line ~146 (the tool table) said `forge_free` "Defaults to the highest quality tier
- (dense geometry + PBR textures)."
-
-The tool table row now reads "Defaults to the standard tier (fast, reliable,
-textured); the caller may request `draft` (fastest) or `high` (best, slower; falls
-back to standard under load)", the tool's own `description` and its `tier` enum
-say the same, and the live evidence for both tiers is in
-`prompts/store-submissions/_generated/forge-free-tier-evidence.json` (a `d97e8e94d`
-attempt to make high the default was rolled back in `b39d6e2f7`).
-
-A reviewer who reads "highest quality tier" and then gets a standard-tier model has
-caught us in a misstatement. Pick the honest resolution and make ALL surfaces agree.
-
-## Your job
-
-Choose based on whether the async high-tier worker can actually ship in this cycle.
-
-### Path A (default, low-risk) — Tell the truth: standard by default
-
-The behavior is fine and defensible (fast, reliable, free). Just make every doc and
-metadata surface say "standard by default, higher tiers available on request."
-
-1. Fix `openai-submission.md` line ~146 (and any other spot) so the `forge_free`
- description matches the code: standard default, `tier` param lets the caller ask
- for draft or high, high may take longer or degrade under load.
-2. Grep every doc for the "highest quality"/"high tier by default" claim and fix each:
- [`docs/mcp-studio.md`](../../docs/mcp-studio.md),
- [`docs/mcp-3d-studio.md`](../../docs/mcp-3d-studio.md), the tool descriptions the
- MCP server itself returns in `tools/list` (check `tools.js` tool metadata strings),
- and `apps-sdk/README.md`.
-3. Make sure the tool's own `description` string returned to ChatGPT is accurate, so
- the model never promises "highest quality" to the user.
-
-### Path B — Make high-by-default real
-
-Only if the async self-host Hunyuan3D worker (`GCP_HUNYUAN3D_URL`) can be deployed
-and proven within this cycle. This is the "never downgrade quality" ideal, and GCP
-GPU spend is pre-approved (`CLAUDE.md` standing resource approvals).
-
-1. Deploy the async Hunyuan3D worker with a real poll handle (submit returns a job id
- immediately, poll returns progress, never blocks 50-280s). Its own
- `workers/<name>/cloudbuild.yaml`, service accounts pinned per `CLAUDE.md`.
-2. Wire `handleForgeFree` to use it for the high tier with a real submit/poll flow
- that a ChatGPT tool call survives, and flip the default to `high`.
-3. Prove it: a real `forge_free` call with no tier returns a high-tier GLB, end to
- end, no timeout, no degrade. Capture the evidence JSON.
-4. THEN the "highest quality by default" doc language becomes true and can stay.
-
-Deploying a worker is not a reason to stop (`CLAUDE.md` self-unblock). But if the
-worker cannot be proven this cycle, do Path A. Do not leave a half-deployed worker
-and a doc claiming it works.
-
-## Constraints
-
-- Every rule in `00-START-HERE.md`.
-- No mock/fake generation. The evidence must be a real GLB from the real lane.
-- This is not a crypto surface; commit gate does not apply.
-
-## Verification
-
-- `grep -rni "highest quality\|high.*by default\|high tier" prompts/store-submissions docs apps-sdk/README.md api/_mcp-studio`
- returns only statements that match the shipped behavior.
-- A real `forge_free` call against `/api/mcp-studio` returns a tier consistent with
- every doc. Save the response JSON as evidence.
-- `npm test` green (run `tests/mcp-forge-free.test.js` and `tests/mcp-studio.test.js`).
- Note which server each covers: `tests/mcp-studio.test.js` is the ChatGPT-facing
- `/api/mcp-studio` tool (standard default). `tests/mcp-forge-free.test.js` is the
- public `mcp-server` package's `forge_free` (`mcp-server/src/tools/forge-free.js`),
- which defaults to `draft` and no longer pins a backend: the forge router's own
- free-lane default (self-host TRELLIS, then Hunyuan3D, then HuggingFace, then NVIDIA
- NIM as the last resort) picks the engine, never the paid Replicate lane.
-
-## Definition of done
-
-- [x] Every doc, the submission sheet, and the tool's own `description` agree with the
- code on the default tier (Path A: standard by default, high opt-in; the async
- Hunyuan3D worker from Path B is deployed and proven, but the default stays
- standard for the reasons in the code comment).
-- [ ] If Path B: worker deployed and proven with real evidence; default is genuinely
- high.
-- [x] If Path A: no surface claims "highest quality by default" anymore.
-- [ ] `npm test` green.
-- [ ] `data/changelog.json` entry only if user-visible behavior changed (Path B yes;
- Path A is a docs/accuracy fix, `docs` tag).
-- [ ] Report states the path taken and includes the real `forge_free` evidence.
diff --git a/docs/openai-pr/04-openai-surface-test-coverage.md b/docs/openai-pr/04-openai-surface-test-coverage.md
deleted file mode 100644
index ff73047a3..000000000
--- a/docs/openai-pr/04-openai-surface-test-coverage.md
+++ /dev/null
@@ -1,107 +0,0 @@
-# Task 04 — Close the test gaps on the ChatGPT-facing surface (P1)
-
-Read [`00-START-HERE.md`](00-START-HERE.md) first.
-
-## The problem
-
-The core logic is well tested (`tests/mcp-studio.test.js`, `tests/ar-export.test.js`,
-`tests/embodiment-*.test.js`), but three surfaces a reviewer will actually hit have
-**no test on the surface itself**, only on the pure functions beneath them. For a
-submission to OpenAI, the HTTP-visible behavior must be pinned so a refactor cannot
-silently break what the reviewer sees.
-
-### Gap 1 — `api/ar.js` HTTP handler is untested
-
-`api/ar.js` is the "View in your space" / "Bring it to life" endpoint every generation
-links to. Only the pure lib `api/_lib/ar-launch.js` was tested (`tests/ar-export.test.js`).
-The handler's own behavior was unverified:
-- User-agent branching (Android → Scene Viewer 302 intent; iOS/desktop → a 200
- interstitial carrying the model's own og:image and title for unfurls, which hands
- off to `/ar/view`, where a real USDZ is generated on the device for Apple Quick
- Look; `kind=avatar` → everyone, Android included, lands on `/ar/view` with the IRL
- handoff).
-- The designed error page for a bad/missing `src`.
-- `assertArAssetUrl` rejection (non-https, non-`.glb/.gltf`) surfacing as a proper
- error response, not a crash.
-- `Vary: user-agent` and cache headers.
-
-Write `tests/api/ar-endpoint.test.js` that invokes the handler with mocked requests
-for each UA class and asserts status, `Location`/redirect, headers, and that the HTML
-launch page contains the model URL and the AR launch elements. Follow the request/
-response mocking pattern used by the other `tests/api/*.test.js` files in this repo.
-
-Closed: [`tests/api/ar-endpoint.test.js`](../../tests/api/ar-endpoint.test.js) covers
-the Android redirect and its `no-store`, the iOS/iPad/desktop interstitial with the
-title and the forwarded host carried through, the `kind=avatar` handoff, the designed
-400 page for every bad-input class (including `.gltf` lookalikes), `Vary: user-agent`,
-the CORS preflight, and an absent User-Agent.
-
-### Gap 2 — `pages/embodiment/embed.html` wiring is untested
-
-The persona widget iframes this page. Its param parsing (`persona/glb/name/state/
-text/emotion/...`), the `GET /api/mcp3d/persona?id=` resolve, and the fallback when a
-persona cannot be resolved are unverified. Add a test (jsdom or Playwright, match what
-the repo already uses for page-level tests) that:
-- Loads the page with `?glb=&name=` and asserts the stage mounts and the name renders.
-- Loads with `?persona=<id>` against a mocked persona endpoint and asserts it resolves
- and renders.
-- Loads with an unresolvable persona and asserts the designed error/empty state, not a
- blank void or console error.
-
-Closed on two levels: [`tests/api/embodiment-embed-page.test.js`](../../tests/api/embodiment-embed-page.test.js)
-pins the page's wiring contract (the documented query params, the persona resolve and
-its failure surface, the designed error when neither `glb` nor `persona` is supplied,
-the opt-in `wallet=1` identity poll, and that a resolved persona's name survives every
-later state change instead of reverting to "Agent"), and
-[`tests/e2e/embodiment-embed.spec.js`](../../tests/e2e/embodiment-embed.spec.js)
-loads the page in a real browser.
-
-### Gap 3 — No contract test binding `api/3d/studio.js` to `openai-actions.yaml`
-
-The custom-GPT Actions response shape (`{status:'done', glbUrl, viewerUrl, arUrl,
-format, previewImageUrl?, tier?}` or `{status:'pending', job, poll, watchUrl,
-previewImageUrl?, tier?, etaSeconds?}`) must match the served OpenAPI schema (Task 05
-serves it). Add a contract test that loads the OpenAPI schema and validates a real
-`api/3d/studio.js` response object against it (the response-shaper is pure and
-exported per the file). Use an existing JSON-schema/OpenAPI validator already in
-`package.json` if present; do not add a heavy new dependency for this.
-
-Closed: [`tests/api/3d-studio-actions-contract.test.js`](../../tests/api/3d-studio-actions-contract.test.js)
-compiles the served
-[`public/.well-known/3d-studio-openapi.yaml`](../../public/.well-known/3d-studio-openapi.yaml)
-with `ajv` (2020-12 draft, plus `ajv-formats`, both already dependencies) and validates
-the exported `shapeSubmit` / `shapePoll` output for the inline-done, queued, running,
-finished, failed, and transient-hiccup states, plus every typed error envelope (400
-`invalid_prompt`, `invalid_tier`, `prompt_rejected`, `bad_request`, `missing_job`,
-`invalid_job`; 429 `rate_limited` with `retry_after`; 502 `generation_failed`; 503
-`not_configured`) against the schema's `GenerationState` and `ErrorResponse`.
-
-## Constraints
-
-- Every rule in `00-START-HERE.md`. No mocked *generation* results that pretend to be
- real GLBs where the test claims to exercise the real lane; where you must stub the
- forge call, stub at the network boundary and label it clearly, and keep at least one
- path that asserts the real shape.
-- Match the existing test style, imports, and helpers. Read a neighboring
- `tests/api/*.test.js` before writing.
-- Not a crypto surface; commit gate does not apply.
-
-## Verification
-
-- `npm test` green, with your three new test files running and passing.
-- Deliberately break each surface locally (e.g. change a header in `api/ar.js`) and
- confirm the new test catches it, then revert. Report that you did this.
-
-## Definition of done
-
-- [x] `tests/api/ar-endpoint.test.js` covers all UA classes, the error page, and
- headers.
-- [x] A page-level test covers `pages/embodiment/embed.html` param parsing, persona
- resolve, and the failure state.
-- [x] A contract test binds `api/3d/studio.js` output to the served OpenAPI (coordinate
- with Task 05 for the schema location).
-- [ ] `npm test` green.
-- [ ] Each new test demonstrably fails when its target is broken (verified, then
- reverted).
-- [ ] No changelog entry needed (tests are internal) unless you also fixed a bug found
- while writing them, in which case that fix gets an entry.
diff --git a/docs/openai-pr/05-custom-gpt-openapi-serving.md b/docs/openai-pr/05-custom-gpt-openapi-serving.md
deleted file mode 100644
index 8750dfc3a..000000000
--- a/docs/openai-pr/05-custom-gpt-openapi-serving.md
+++ /dev/null
@@ -1,82 +0,0 @@
-# Task 05 — Serve the custom-GPT OpenAPI and align legal URLs (P1)
-
-Read [`00-START-HERE.md`](00-START-HERE.md) first. Coordinate with Task 02 (the
-served schema must be payment-free) and Task 04 (a contract test binds to it).
-
-## The problem (verified, with evidence)
-
-The custom-GPT Actions schema
-([`prompts/store-submissions/_generated/openai-actions.yaml`](../../prompts/store-submissions/_generated/openai-actions.yaml),
-OpenAPI 3.1.0, `title: three.ws 3D Studio`) exists but is **not served anywhere
-machine-discoverable**. It lives only under `prompts/store-submissions/_generated/`.
-
-Meanwhile the served `public/.well-known/openapi.yaml` describes the general REST
-platform and does NOT include `/api/3d/studio`, `/api/ar`, or `forge_free`. So the
-machine-readable schema an OpenAI reviewer or the GPT Actions runtime can fetch does
-not describe the 3D Studio GPT at all.
-
-Additionally, the legal URLs inside `openai-actions.yaml` use `.html` suffixes
-(`https://three.ws/legal/privacy.html`, `.../legal/tos.html`) while the submission doc
-verifies the live URLs without `.html`. Both forms currently resolve (verified:
-`vercel.json` rewrites `/legal/privacy/?` to `/legal/privacy.html`, and the `.html`
-file is served directly), but the inconsistency is sloppy for a submission and should
-be normalized to one canonical form.
-
-## Your job
-
-1. **Serve the 3D Studio OpenAPI at a stable, discoverable path.** Put the served
- copy where the GPT Action and a reviewer can fetch it over https, e.g.
- `public/.well-known/3d-studio-openapi.yaml` (or a `/api/3d/openapi.yaml` route).
- Pick the path that fits how this repo already serves the other `.well-known`
- schemas and wire it in `vercel.json` if a route is needed. Do NOT overwrite the
- platform's existing `public/.well-known/openapi.yaml`.
-
-2. **Make the served 3D Studio schema self-contained and payment-free.** It must
- describe only `/api/3d/studio` (POST submit, GET poll) and `/api/ar`, with
- `auth: none`, no x402, no prices, no chain references. This is the artifact Task 02
- points at as the app's clean discovery surface.
-
-3. **Single source of truth.** The served schema and
- `prompts/store-submissions/_generated/openai-actions.yaml` must not drift. Either
- generate the `_generated` copy from the served one (preferred, matches the
- `_generated/` build-script convention already in that dir) or make one a copy of
- the other with a script/check. Do not hand-maintain two copies.
-
-4. **Normalize the legal URLs.** Pick the canonical form the rest of the submission
- uses (the no-`.html` form, e.g. `https://three.ws/legal/privacy`,
- `https://three.ws/legal/tos`) and use it consistently in the served schema, the
- `_generated` copy, and `openai-submission.md`. Confirm every legal URL you cite
- returns 200 (both `privacy` and `tos` at minimum).
-
-5. **Lint it.** `npx @redocly/cli lint <served-schema>` must pass clean (the TRACKER
- claims the `_generated` copy already lints clean; keep it that way for the served
- one).
-
-## Constraints
-
-- Every rule in `00-START-HERE.md`. The served schema must pass Task 02's
- no-crypto-tokens test, so keep it strictly the two free endpoints.
-- Not a crypto surface if you keep it clean; just do not paste x402 content into it.
-- If you add a `vercel.json` route, remember `vercel.json` is a LIVE config consumed
- by the server (`CLAUDE.md`); edit it carefully and keep the route table valid.
-
-## Verification
-
-- The served schema is fetchable (`curl` the local dev server or read the built
- `dist/` copy) and validates with `@redocly/cli lint`.
-- Every legal URL in the schema returns 200 (test both privacy and tos).
-- Task 04's contract test validates a real `api/3d/studio.js` response against it.
-- `npm run build:pages` / the normal build still succeeds; `npm test` green.
-
-## Definition of done
-
-- [ ] The 3D Studio OpenAPI is served at a stable https path, describing only the two
- free endpoints, `auth: none`, no payment/crypto content.
-- [ ] The served copy and the `_generated` copy are kept in sync by a script or check,
- not by hand.
-- [ ] All legal URLs normalized to one canonical form and verified 200.
-- [ ] Schema lints clean.
-- [ ] `npm test` green; build succeeds.
-- [ ] `data/changelog.json` entry (`docs` or `infra` tag) since a new public schema
- path is developer-visible; add the route to `data/pages.json` only if it is a
- human page (a raw schema usually is not).
diff --git a/docs/partners.md b/docs/partners.md
index 527ec8c6b..cea50c5bf 100644
--- a/docs/partners.md
+++ b/docs/partners.md
@@ -37,7 +37,7 @@ The page's stat bar reads 8 partners, 22 chains, and an infinity glyph for scale
- **AR handoff** works from any of these: every generation carries a place-in-your-room link, and the whole pipeline is public and keyless rather than ChatGPT-exclusive. See [AR in ChatGPT](./chatgpt-ar.md).
- **Spatial MCP** is the open, CC0 response shape that makes a 3D scene a native MCP result instead of a URL in text. three.ws is the reference implementation; the shape is renderer-agnostic and carries no payment, wallet, or coin surface. See [Spatial MCP](./spatial-mcp.md).
-**Saying it correctly.** Write the status as "OpenAI Select Partner". three.ws is an independent member of the network at the Select tier: not an OpenAI product, and not endorsed by OpenAI beyond the partner designation. Badge rules, the required independence line, and every asset are in [`marketing/openai-select-partner/badge-usage.md`](../marketing/openai-select-partner/badge-usage.md); the graphics cleared for editorial use are in [Press kit](./press-kit.md). The pre-submission verification pack is [`docs/openai-pr/00-START-HERE.md`](./openai-pr/00-START-HERE.md).
+**Saying it correctly.** Write the status as "OpenAI Select Partner". three.ws is an independent member of the network at the Select tier: not an OpenAI product, and not endorsed by OpenAI beyond the partner designation. Badge rules, the required independence line, and every asset are in [`marketing/openai-select-partner/badge-usage.md`](../marketing/openai-select-partner/badge-usage.md); the graphics cleared for editorial use are in [Press kit](./press-kit.md). The pre-submission verification pack is [`prompts/openai-pr/00-START-HERE.md`](../prompts/openai-pr/00-START-HERE.md).
---
diff --git a/marketing/openai-select-partner/partnership-and-gpt-briefing.md b/marketing/openai-select-partner/partnership-and-gpt-briefing.md
index 295505835..bfcf40537 100644
--- a/marketing/openai-select-partner/partnership-and-gpt-briefing.md
+++ b/marketing/openai-select-partner/partnership-and-gpt-briefing.md
@@ -419,7 +419,7 @@ that is now wrong. Fix all three before the release is emailed for approval.
| Partner ecosystem overview | `docs/partners.md` |
| App Directory answer sheet and audit | `prompts/store-submissions/_generated/openai-submission.md` |
| Submission status tracker | `prompts/store-submissions/_generated/TRACKER.md` |
-| Apps SDK handoff pack | `docs/openai-pr/00-START-HERE.md` |
+| Apps SDK handoff pack | `prompts/openai-pr/00-START-HERE.md` |
| Actions endpoint | `api/3d/studio.js` |
| MCP connector | `api/mcp-studio.js`, `api/_mcp-studio/*` |
| Inline widget | `api/_mcp-studio/component.js` |
diff --git a/prompts/openai-pr/00-START-HERE.md b/prompts/openai-pr/00-START-HERE.md
new file mode 100644
index 000000000..ec402beab
--- /dev/null
+++ b/prompts/openai-pr/00-START-HERE.md
@@ -0,0 +1,94 @@
+# OpenAI Apps SDK submission: agent handoff pack
+
+This pack holds the task briefs for submitting the three.ws **3D Studio** app to OpenAI
+(ChatGPT Apps SDK / App Directory). It moved here from `docs/openai-pr/` on 2026-09-01 so
+that every open work order in the repo lives under [prompts/](../README.md) and follows its
+standard. Each `NN-*.md` file is a complete prompt for one agent. Read this file first, then
+the brief for your task.
+
+## What we are shipping
+
+The submission target is the **free, keyless MCP connector** at
+`https://three.ws/api/mcp-studio` plus its inline widget and the custom-GPT Actions surface.
+A reviewer at OpenAI will exercise these surfaces and read the manifests we expose.
+Everything they can reach must be correct, consistent, and free of any paid or crypto surface.
+
+The submission kit lives at [`prompts/store-submissions/_generated/`](../store-submissions/_generated/)
+(answer sheet `openai-submission.md`, `openai-actions.yaml`, `TRACKER.md`, screenshots,
+evidence JSON). Your job is not to recreate it. Your job is to close the specific gaps that
+would fail review or embarrass us in front of OpenAI.
+
+## The map of the surface
+
+| Surface | Where it lives | Notes |
+| --- | --- | --- |
+| Free MCP connector (submission target) | `api/mcp-studio.js`, `api/_mcp-studio/*` | Keyless. Live `tools/list` returns 11 tools (8 in `tools.js` plus 3 persona tools; `look_at_model` was added 2026-08-28). Real forge lane. |
+| Inline ChatGPT widget | `api/_mcp-studio/component.js` | `<model-viewer>` skybridge resource `ui://widget/three-studio-model.html` |
+| Embodiment embed | `apps-sdk/embodiment/*`, `pages/embodiment/embed.html` | Persona widget iframes this. |
+| AR-in-ChatGPT | `api/ar.js`, `api/_lib/ar-launch.js` | Every generation carries `arUrl`. |
+| Custom GPT Actions (REST) | `api/3d/studio.js`, `prompts/store-submissions/_generated/openai-actions.yaml` | GPT Store listing is live. Served copy: `public/.well-known/3d-studio-openapi.yaml`, byte-pinned by `npm run check:studio-openapi`. |
+| Served manifests | `public/.well-known/*` | Publicly discoverable. |
+| Paid MCP server (NOT the submission) | `api/mcp-3d.js`, `api/_mcp3d/*` | x402/OAuth gated. Out of scope unless a task says so. |
+
+## The tasks
+
+Briefs 01 through 05 were verified shipped on 2026-09-01 and deleted; they remain readable in
+git history (`git log --diff-filter=D -- docs/openai-pr/`). What each closed:
+
+| # | Brief | Verified by |
+| --- | --- | --- |
+| 01 | Orphaned studio-viewer widget removed | `apps-sdk/studio-viewer`, `scripts/build-apps-sdk-viewer.mjs`, `public/apps-sdk` gone since `19c379125` (2026-07-18) |
+| 02 | `.well-known` manifests reconciled with the zero-payment claim | `ai-plugin.json` leads with the free lane and a branded logo; `tests/wellknown-manifests.test.js` asserts the free schema is crypto-free; served and kit copies byte-identical |
+| 03 | `forge_free` tier story true everywhere | `api/_mcp-studio/tools.js` defaults to `standard`; the live tool description, the answer sheet and `forge-free-tier-evidence.json` agree |
+| 04 | Test coverage on the ChatGPT-facing surface | `tests/api/ar-endpoint.test.js` (20), `tests/api/embodiment-embed-page.test.js` (7), `tests/api/3d-studio-actions-contract.test.js`, `tests/e2e/embodiment-embed.spec.js` (6); production `/api/ar` matches per user agent |
+| 05 | Custom-GPT OpenAPI served, legal URLs aligned | `https://three.ws/.well-known/3d-studio-openapi.yaml` 200, two paths, `security: []`, Redocly lint clean; `/legal/privacy` and `/legal/tos` 200 |
+
+Open:
+
+| # | Brief | Priority | State (measured 2026-09-01) |
+| --- | --- | --- | --- |
+| 06 | [Fix doc accuracy and reconcile the tool-count story](06-docs-accuracy-reconciliation.md) | P1 | Partial. The 2026-08-06 reconciliation shipped and `docs/mcp-studio.md` says eleven, but `api/_mcp-studio/tools.js:3`, `openai-submission.md`, `live-tools-list.json`, `TRACKER.md`, the `/openai` page copy and its locale key, and the partnership brief still say ten or nine. |
+| 07 | [Final live verification and submit checklist](07-final-verification-and-submit.md) | P2 (runs last) | Never run. Its "exactly 10 tools" check must read 11; `initialize`, `tools/list`, `resources/list`, the served manifests and `/api/ar` pass today; the live `forge_free` and `POST /api/3d/studio` runs and the kit refresh remain. |
+
+Run 06 first, then 07. The human steps at the end of 07 (partner-portal submit, deleting the
+draft GPT) stay with the owner.
+
+## Rules every task must follow (from `CLAUDE.md`)
+
+These are non-negotiable and OVERRIDE any default behavior:
+
+1. **No mocks, no fake data, no placeholders, no TODOs, no stubs, no commented-out
+ code.** If you write it, finish it and wire it. Real APIs only.
+2. **Execute, do not interview.** Pick the most reasonable interpretation, do it,
+ verify it, and report. Do not end your turn with "want me to proceed?".
+3. **Every state is designed** (loading/empty/error/populated) for any UI you touch.
+4. **Tests must pass:** run `npm test` (do NOT pipe through `tail`; it masks the
+ exit code; a vitest failure gates the Playwright stage).
+5. **Changelog:** any user-visible change gets an entry in
+ [`data/changelog.json`](../../data/changelog.json) (date, plain-language title +
+ summary, tags from feature/improvement/fix/sdk/infra/docs/security). Run
+ `npm run build:pages` to regenerate and validate the feed. Internal-only chores
+ do not get an entry.
+6. **Docs ship with the feature.** If you touch a surface and its doc is now wrong,
+ fix the doc in the same change. New page/route: `data/pages.json`.
+7. **Commit gate.** `$THREE` (`FeMbDoX7R1Psc4GEcvJdsbNbZA3bfztcyDCatJVJpump`) is the
+ only coin this platform promotes. Before any `git commit` whose diff references ANY
+ crypto project other than `$THREE` (in code, docs, manifests, tests, or commit text),
+ STOP and get explicit owner approval. Do the work, stage nothing that trips the gate,
+ and flag it.
+8. **Do not push, deploy, or submit to OpenAI without owner approval.** Commit
+ locally, prepare everything so the ship is one command, and say what remains.
+9. **Never use the em-dash character.** Use a period, comma, colon, or parentheses.
+10. **Concurrent agents share this worktree.** Stage explicit paths only (never
+ `git add -A`). Re-check `git status` and `git diff --staged` right before you
+ commit.
+
+## Definition of done for the whole pack
+
+- [ ] 06 and 07's own definitions of done are met.
+- [ ] `npm test` is green.
+- [ ] No surface a reviewer can reach advertises a paid or crypto capability.
+- [ ] Every doc and manifest describing the ChatGPT surface matches the running code,
+ including the tool count.
+- [ ] The only remaining steps are the human `[HUMAN: ...]` items in
+ [`TRACKER.md`](../store-submissions/_generated/TRACKER.md).
diff --git a/docs/openai-pr/06-docs-accuracy-reconciliation.md b/prompts/openai-pr/06-docs-accuracy-reconciliation.md
similarity index 96%
rename from docs/openai-pr/06-docs-accuracy-reconciliation.md
rename to prompts/openai-pr/06-docs-accuracy-reconciliation.md
index 6a307e240..b3b5f421e 100644
--- a/docs/openai-pr/06-docs-accuracy-reconciliation.md
+++ b/prompts/openai-pr/06-docs-accuracy-reconciliation.md
@@ -1,4 +1,4 @@
-# Task 06 — Fix doc accuracy and reconcile the tool-count story (P1)
+# Task 06: Fix doc accuracy and reconcile the tool-count story (P1)
Read [`00-START-HERE.md`](00-START-HERE.md) first. This task cleans up the
documentation drift the audit found, EXCEPT the two docs owned by other tasks
@@ -18,7 +18,7 @@ claim is Task 03). Do not fight those tasks for those specific edits; coordinate
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
+ 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
diff --git a/docs/openai-pr/07-final-verification-and-submit.md b/prompts/openai-pr/07-final-verification-and-submit.md
similarity index 97%
rename from docs/openai-pr/07-final-verification-and-submit.md
rename to prompts/openai-pr/07-final-verification-and-submit.md
index d3bcb786d..ce23e3a3c 100644
--- a/docs/openai-pr/07-final-verification-and-submit.md
+++ b/prompts/openai-pr/07-final-verification-and-submit.md
@@ -1,4 +1,4 @@
-# Task 07 — Final live verification and submit checklist (P2, runs last)
+# Task 07: Final live verification and submit checklist (P2, runs last)
Read [`00-START-HERE.md`](00-START-HERE.md) first. **Run this only after tasks
01-06 are merged.** This is the go/no-go gate before the app is submitted to OpenAI.
@@ -6,7 +6,7 @@ Read [`00-START-HERE.md`](00-START-HERE.md) first. **Run this only after tasks
## Purpose
The submission answer sheet
-([`prompts/store-submissions/_generated/openai-submission.md`](../../prompts/store-submissions/_generated/openai-submission.md))
+([`prompts/store-submissions/_generated/openai-submission.md`](../store-submissions/_generated/openai-submission.md))
asserts many "verified live" facts, dated 2026-07-14. Deployments move. Before we put
this in front of OpenAI, re-verify every claim against the *currently deployed*
production, close the loose `[HUMAN: ...]` items where you can, and produce a clean