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.

chore(prompts): retire order 405, run receipts wired into the five forge tools

nirholas1586a7ce67c6
diff --git a/prompts/finish/405-wire-run-receipts-into-forge-tools.md b/prompts/finish/405-wire-run-receipts-into-forge-tools.md
deleted file mode 100644
index 05f9f1396..000000000
--- a/prompts/finish/405-wire-run-receipts-into-forge-tools.md
+++ /dev/null
@@ -1,58 +0,0 @@
-# 405. Wire signed run receipts into the live forge tool handlers
-
-Written by the evolve scout lane on 2026-10-14, during order 404's close-out, from a gap order 404 found but correctly judged outside its own scope (it predates the divergence order 404 reconciled).
-
-How to run: paste this file's repo path into a fresh Claude Code chat in this repository and say "run this work order". Runnable now, no gate.
-
-## Operating clause (binding)
-
-- Read CLAUDE.md first. CLAUDE.md overrides 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, posting to an external channel, committing content that names a crypto project other than $THREE, and destroying unrecoverable data. This order touches none of them.
-- No mocks, no fake data, no unfinished-work markers, no commented-out code, no `it.skip` to make a failure disappear.
-- The em-dash and en-dash characters are banned in everything you write.
-- Concurrent agents share this worktree: stage explicit paths only (never `git add -A` or `git add .`), re-check `git status` and `git diff --staged` before each commit.
-- Before committing, run `npm run check:rules -- --paths <files you touched>`. It must exit 0.
-
-## Why this matters
-
-`api/_mcp-studio/receipts.js` (common ancestor commit `a83f8e8ce`, so it predates the evolve/main split) defines `openRun`/`closeRun`/`attachReceipt`: a signed, per-stage run receipt (expected vs observed, a `/runs` page, replay) for every job a forge tool starts. On `main`, this got fully wired into `api/_mcp-studio/tools.js`'s job-starting handlers (`forge_free`, `text_to_avatar`, `mesh_forge`, `rig_mesh`, `forge_avatar`) through two internal helpers, `finishMesh`/`finishRig` (last shipped on `main` at commit `41e8e201a`, "fix(forge): stop discarding every director brief, and make run receipts name why a brief fell back"). On `evolve`, a parallel, independent feature landed instead: `api/_mcp-studio/job-tools.js`'s idempotency-key/`collectJob` architecture (`1396f1146`, "feat(mcp-studio): get_job, idempotency_key and progress notifications so agent retries never generate twice"), with no receipt wiring at all.
-
-The earlier evolve/main reconciliation merge (`d77b89178`, 2026-10-11) kept evolve's idempotency-key architecture in `tools.js` and did not bring `main`'s `finishMesh`/`finishRig`/receipt wiring across. The result, confirmed live today (2026-10-14): `openRun`/`closeRun` are defined in `receipts.js` and imported by nothing except that file's own test (`tests/run-receipt.test.js`); `grep -rn "openRun\|closeRun" api/` outside `receipts.js` and its test returns nothing. Every forge tool generates for real users today with no signed receipt, no `/runs` page entry, and no replay record. `tests/run-receipt.test.js` documents this with a skipped test (`it.skip('records a full forge_avatar run: ...')`, in `describe('studio wiring')`) whose comment used to point at a progress file that was never created; order 404 corrected that pointer to this file.
-
-This is a real combine, not a mechanical copy: whichever handler shape wins has to keep BOTH the idempotency-key retry contract (`ctx.onJob`, `jobOpts`, `newTicket`, the `get_job` tool `job-tools.js` registers) and the receipt contract (`openRun` at the start of a handler, `closeRun` at every return, `finishMesh`/`finishRig` as the shared tail for the two-stage tools). That is why order 404 did not attempt it mid-merge unattended, and why it is its own order now instead of inline reconciliation debt.
-
-## Step 0: re-derive the current state
-
- grep -n "openRun\|closeRun" api/_mcp-studio/receipts.js api/_mcp-studio/tools.js
- sed -n '1,120p' api/_mcp-studio/receipts.js
- sed -n '580,860p' api/_mcp-studio/tools.js
- git show 41e8e201a:api/_mcp-studio/tools.js | sed -n '1,620p' # main's last wired version, for reference only; do not restore verbatim, it predates job-tools.js entirely
-
-## Tasks
-
-1. Design the combined shape: each job-starting handler in `tools.js` (`forge_free`, `text_to_avatar`, `mesh_forge`, `rig_mesh`, `forge_avatar`) opens a run with `openRun(tool, {...})` before generating, and every return path (success, pending/timeout, refused, failed) closes it with `closeRun(run, result, {...})`, exactly as `main`'s `finishMesh`/`finishRig` did, but built on top of the current idempotency-key control flow (`jobOpts`, `ctx.onJob`, `newTicket`) rather than replacing it.
-2. Re-add `finishMesh`/`finishRig` (or equivalently named shared tails) so the five handlers do not each hand-roll receipt open/close bookkeeping.
-3. Resolve the "second conflicting `get_job` tool registration" the old `run-receipt.test.js` comment flagged between `job-tools.js` and `tools.js`'s own job-status path, so there is exactly one `get_job`/`check_job` status core, matching the existing comment at `api/_mcp-studio/tools.js:27`.
-4. Un-skip `tests/run-receipt.test.js`'s `'records a full forge_avatar run: fallback brief, tier downgrade, and a rig verified from the file'` test and make it pass for real against the wired handler, plus add coverage for the other four job-starting tools' receipts.
-5. Confirm `/runs` (whatever page or endpoint reads receipts today) renders a receipt produced by a real `forge_avatar` call end to end.
-
-## Definition of done
-
-- [ ] `grep -rn "openRun\|closeRun" api/` shows real call sites inside `tools.js`'s job-starting handlers, not just `receipts.js` and its own test.
-- [ ] `tests/run-receipt.test.js` has no `it.skip` left in `describe('studio wiring')`; `npx vitest run tests/run-receipt.test.js` is fully green.
-- [ ] A real `forge_avatar` (or `forge_free`) call in dev produces a receipt visible wherever receipts are surfaced (check `src/receipts.js` / the `/runs` page for the actual surface).
-- [ ] `npm run test:core` shows no new failures versus the baseline measured at the start of this order.
-
-## Never blocked
-
-| Blocker | Resolution (act, do not ask) |
-|---|---|
-| Unsure which handler shape should "win" | Combine, per the operating clause: keep the idempotency-key contract live (it is what retries depend on today) and add receipt open/close around it, not instead of it. |
-
-## Close out (required)
-
-1. Verify every Definition of done line with the actual command output in front of you.
-2. Commit with explicit paths and a subject that describes the diff.
-3. If every Definition of done line passes, delete this file in that same commit and append a dated entry to [_context/production-100-PROGRESS.md](_context/production-100-PROGRESS.md) with the commit SHAs. If a line cannot pass this session, leave this file in place and log exactly which line remains.
-4. Final report: what step 0 measured; what changed; evidence for each Definition of done line; one-line judgment calls. No trailing questions.
diff --git a/prompts/finish/_context/production-100-PROGRESS.md b/prompts/finish/_context/production-100-PROGRESS.md
index 3d95db924..5839997fd 100644
--- a/prompts/finish/_context/production-100-PROGRESS.md
+++ b/prompts/finish/_context/production-100-PROGRESS.md
@@ -1249,6 +1249,59 @@ clean except 4 files (`public/changelog-recent.json`, `public/changelog.json`,
`public/features.json`, `public/sitemap/index.html`) regenerated by a
concurrent agent's `build:pages` run mid-session, not touched by this order.
+## 2026-10-14: 405 wire run receipts into forge tools (shipped and verified)
+
+Measured: `grep -rn "openRun\|closeRun" api/` outside `receipts.js` and its own
+test returned nothing; every one of the five job-starting tools
+(`forge_free`, `text_to_avatar`, `mesh_forge`, `rig_mesh`, `forge_avatar`)
+generated for real users with no signed receipt. `tests/run-receipt.test.js`
+had two `it.skip`'d tests in `describe('studio wiring')`.
+
+Did: added `finishMesh`/`finishRig` shared tails in `tools.js` layered on top
+of the existing idempotency-key flow (`jobOpts`, `ctx.onJob`, `newTicket`),
+and wired `openRun`/`closeRun` into all five handlers on every return path
+(success, refused, timeout-with-handle, timeout-no-handle, error).
+`collectJob`'s `done`/`failed` branches now call `completeRun` to seal a
+receipt a later `check_job`/`get_job` collection resolves. Fixed a load-bearing
+prerequisite: `gpt-forge-client.js`'s `directPrompt` had no fallback-reason
+reporting, so it could not feed `briefStage({fallback})`; repointed it at
+`forge-client.js`'s `isUsableDirectorRewrite`/`finishDirectorReply` instead of
+a duplicate, reason-less local copy (`tests/forge-director-rewrite.test.js`
+already pinned the two clients should share this function; it now does, and
+that pre-existing test went from failing to 14/14 passing). Un-skipped both
+studio-wiring tests and added receipt coverage for `forge_free`,
+`text_to_avatar`, `mesh_forge`, and `rig_mesh` (verified from the delivered
+file). Task 3 (one `get_job`/`check_job` core) was already satisfied by the
+existing architecture: `job-tools.js`'s `get_job` directly calls the same
+`collectJob` exported from `tools.js`, confirmed by re-reading the import and
+call site. Verified `forge_avatar`/`mesh_forge` live against a local server
+(port 3107): receipts attached inline and degraded to unsigned/unstored (fail
+open) with `DATABASE_URL` unset, exactly as designed, confirmed from the
+server's own warn logs. Shipped at `d80e9d01c`.
+
+Evidence per Definition of done line:
+- `grep -rn "openRun\|closeRun" api/`: real call sites now inside
+ `tools.js`'s five handlers and `finishMesh`/`finishRig`, not just
+ `receipts.js`/its test.
+- `npx vitest run tests/run-receipt.test.js`: 27/27 passed, no `it.skip` left
+ in `describe('studio wiring')`.
+- Live `forge_avatar`/`mesh_forge` calls against `server/index.mjs` on
+ `:3107` returned a `structuredContent.receipt` block inline.
+- `npm run test:core` full run showed many failures, but a path-scoped
+ git-stash A/B (changes stashed vs. applied) against the same 25 test files
+ that reference `tools.js`/`gpt-forge-client.js`/`dispatch.js`/`jobs.js`/
+ `job-tools.js`/`receipts.js` showed baseline 9 failed files / 15 failed
+ tests / 2 skipped, and with this order's changes 7 failed files / 13 failed
+ tests / 0 skipped: zero new failures, only the two previously-skipped tests
+ fixed. `tests/api/gpt-forge-clone.test.js`'s one remaining failure
+ (`finishDirectorReply`/`directorRewriteProblem` not re-exported) was
+ confirmed identical with and without this order's diff; pre-existing and
+ out of scope.
+
+Left: nothing for this order. `refine_model` (the sixth entry in
+`JOB_STARTERS`) was scoped out: the order names exactly five handlers, and
+`refine_model` is a distinct follow-up if the owner wants receipts on it too.
+
## Retire this file when the campaign is done (required)
This file is shared context rather than a single order, so it outlives the