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(roadmap): refresh the reuse map and record the gate baselines for the latest audit runs

nirholasf5e9be4e2a68
diff --git a/prompts/README.md b/prompts/README.md
index 13a1ecd4e..063a5abe6 100644
--- a/prompts/README.md
+++ b/prompts/README.md
@@ -66,7 +66,7 @@ state measured on 2026-09-01; trust the directory, never this table.
| `openai-pr-` | The OpenAI Apps SDK submission pack. Briefs 01 to 05 shipped; 06 (the tool count drifted to 11 across the kit) and 07 (the go/no-go, never run) remain, with the portal submit owner-only. |
| `production-100-` | The run-to-100% campaign: a master index sequencing every open work order toward a mechanically checkable "production ready, roadmap complete" end state, a standing ship-readiness order, closeout orders for residuals no other campaign owned (stranded custodial funds, master-key hygiene, verdict tuning), and the batched owner-action list. Its map was re-measured against code and production on 2026-09-01; start there. |
| `quality-bar-` | The GCP-credit quality campaign. Six open orders (fleet scale, PBR materials, forge UX, design system, mobile, avatar likeness); the reference pipeline, flagship lane, viewers and eval harness shipped. |
-| `roadmap-` | Five runnable orders for existing surfaces (generation suite, creation consolidation, parametric avatar editor, developer resources, native home-screen widgets), plus the strategy layer ([the playbook](finish/roadmap-fable-playbook.md)) that decides what to run next and [the reuse map](finish/roadmap-REUSE-MAP.md) for license-vetted OSS. |
+| `roadmap-` | Five runnable orders for existing surfaces (generation suite, creation consolidation, parametric avatar editor, developer resources, native home-screen widgets), plus the strategy layer ([the playbook](../docs/internal/fable-playbook.md)) that decides what to run next and [the reuse map](finish/roadmap-REUSE-MAP.md) for license-vetted OSS. |
| `simulation-ready-` | The physics-grade asset campaign: turning generated 3D into assets a rigid-body simulator can consume unedited (metric scale, watertightness, mass and inertia, a collision proxy), signed into the provenance credential and filterable by a buying agent. Its context file carries the frontier bet, the scored candidate table it beat, and the measured kernel evidence (20 live assets graded; 2 of 20 usable as-is, 0 of 10 from our own lanes). Awaiting its architect pass. |
| `store-submissions-` | Listing three.ws MCP tools across the Claude and OpenAI marketplaces and the MCP registries. All 21 numbered orders shipped; [the closeout](finish/store-submissions-01-submission-closeout.md) owns the remaining code gaps and the human submission steps. |
| `swarm-100-` | The "100% production ready, roadmap complete" goal decomposed into fully independent, single-task orders (per-route browser audits, per-batch api/cron/docs audits, per-worker and per-SDK audits, repo-wide sweeps, and one order per README-roadmap slice). Any file runs standalone in a fresh chat and depends on no index (157 left on 2026-09-01: 151 route audits, 4 sweeps, 1 roadmap slice). |
diff --git a/prompts/finish/production-100-00-INDEX.md b/prompts/finish/production-100-00-INDEX.md
index fc32eef03..b5b76d879 100644
--- a/prompts/finish/production-100-00-INDEX.md
+++ b/prompts/finish/production-100-00-INDEX.md
@@ -100,7 +100,7 @@ run in parallel, H waits on owner gates, I is anytime, J is the parallel swarm.
| [../simulation-ready/01-architecture.md](simulation-ready-01-architecture.md) | Kernel, spec and schema exist; 0.5 of 8 build tasks done. |
| Trading trio (see [../roadmap/00-README.md](roadmap-00-README.md)) | Three plans, largely absorbed by shipped surfaces; every commit touching them is gated. |
-Strategy for what to run next inside E: [../roadmap/fable-playbook.md](roadmap-fable-playbook.md).
+Strategy for what to run next inside E: [the Fable playbook](../../docs/internal/fable-playbook.md).
### F. Trust and benchmarks
diff --git a/prompts/finish/roadmap-00-README.md b/prompts/finish/roadmap-00-README.md
index d81cf33be..27b5fadd4 100644
--- a/prompts/finish/roadmap-00-README.md
+++ b/prompts/finish/roadmap-00-README.md
@@ -6,7 +6,7 @@ Each file here is a **self-contained prompt** you paste into a fresh Claude Code
Read this file once; every prompt assumes it.
-**Strategy layer:** [fable-playbook.md](roadmap-fable-playbook.md) is the operating strategy
+**Strategy layer:** [the Fable playbook](../../docs/internal/fable-playbook.md) is the operating strategy
above these prompts — how to deploy Claude Fable 5 across them, the revenue ladder,
OSS in/out motions, and standing routines. Read it to decide *what to run next*;
read this file to run it safely.
@@ -81,7 +81,7 @@ retired; they remain readable in git history
| File | What it is |
|---|---|
-| [fable-playbook.md](roadmap-fable-playbook.md) | The operating strategy above these work orders: what to run next, the revenue ladder, standing routines. |
+| [the Fable playbook](../../docs/internal/fable-playbook.md) | The operating strategy above these work orders: what to run next, the revenue ladder, standing routines. Promoted out of this directory into `docs/internal/` on 2026-09-03, since it is durable strategy rather than a one-shot order. |
| [REUSE-MAP.md](roadmap-REUSE-MAP.md) | License-vetted OSS to integrate instead of reinventing. Check it before building anything here. |
| [pumpfun-trading.md](roadmap-pumpfun-trading.md), [pumpfun-trading-arena.md](roadmap-pumpfun-trading-arena.md), [pumpfun-trading-wedge.md](roadmap-pumpfun-trading-wedge.md) | Three overlapping strategy documents for the trading product. They are plans, not work orders, and every commit touching them hits the CLAUDE.md commit gate. The wedge document's own recommendation stands: fold the arena draft into the master plan, keep the wedge as the delta. |
diff --git a/prompts/finish/roadmap-REUSE-MAP.md b/prompts/finish/roadmap-REUSE-MAP.md
index 9c4f3c21f..2c40114a4 100644
--- a/prompts/finish/roadmap-REUSE-MAP.md
+++ b/prompts/finish/roadmap-REUSE-MAP.md
@@ -1,112 +1,152 @@
-# OSS Reuse Map — integration targets for the 3D/crypto/AI roadmap
+# OSS Reuse Map: integration targets for the 3D/crypto/AI roadmap
-Verified June 2026. Use this before building anything in `prompts/roadmap/*` — prefer integrating these proven, permissively-licensed projects over reinventing.
+License survey June 2026. Integration status re-verified against the working
+tree on 2026-09-03 (see "How this file was verified" at the bottom). Use this
+before building anything in `prompts/roadmap/*`: prefer integrating these
+proven, permissively-licensed projects over reinventing.
-**License legend:** ✅ permissive (MIT/Apache-2.0/BSD/Zlib — safe to ship) · ⚠️ conditional (revenue/MAU/territory caps — read the fine print) · ⛔ AVOID (non-commercial / research / GPL/AGPL / unlicensed).
+**License legend:** ✅ permissive (MIT/Apache-2.0/BSD/Zlib, safe to ship) · ⚠️ conditional (revenue/MAU/territory caps, read the fine print) · ⛔ AVOID (non-commercial / research / GPL/AGPL / unlicensed).
-**Vercel reality:** no GPU model runs *inside* a Vercel function. The pattern everywhere: Vercel function = thin orchestrator that POSTs to a GPU host (Replicate/HF/Modal/Cloud Run) and polls/webhooks for the result URL. This matches the existing Forge architecture.
+**Status legend:** **[wired]** already integrated here, do not re-adopt · **[dep-only]** in `package.json` but imported nowhere · **[open]** not adopted, still a live option.
+
+## Runtime reality (read before you design)
+
+Production runs on **Google Cloud Run**, not Vercel (migrated 2026-07-07; see
+CLAUDE.md "Stack notes"). One container serves the static frontend, the
+`vercel.json` route table, and every `api/**` handler. Earlier revisions of this
+file framed every verdict as "can this run in a Vercel function"; that framing
+is retired, and any advice below that survives it is restated in Cloud Run terms.
+
+What did not change is the GPU rule: no GPU model runs inside the API container.
+The pattern everywhere is that an `api/` handler is a thin orchestrator that
+POSTs to a GPU host (a `workers/<name>` Cloud Run service, Replicate, HF, Modal)
+and polls or takes a webhook for the result URL. This matches the existing Forge
+architecture. What Cloud Run buys you over the old Vercel constraint is that a
+native binary or a headless chromium in the image is now a normal option rather
+than a disqualifier, so "needs a native dep" is no longer on its own a reason to
+reject a library.
---
-## 1. GLB/glTF compression (Draco + meshopt) — for roadmap 02, 04
-| Capability | Repo / npm | License | Note |
-|---|---|---|---|
-| Build-time pipeline (primary) | donmccurdy/glTF-Transform · `@gltf-transform/cli`+`/core`+`/functions` | ✅ MIT | `gltf-transform optimize in.glb out.glb --compress meshopt`. Use as build/Vercel step. |
-| Draco runtime decode | three `DRACOLoader` + `draco3d` (google/draco) | ✅ Apache-2.0 | Copy `three/examples/jsm/libs/draco/` to `public/`; `loader.setDRACOLoader`. Max geometry compression. |
-| Meshopt runtime decode (prefer) | zeux/meshoptimizer · `meshoptimizer`,`gltfpack` | ✅ MIT | `loader.setMeshoptDecoder(MeshoptDecoder)`. Tiny + fast decode → best runtime default. |
-
-## 2. Web AR + GLB→USDZ (iOS Quick Look) — for roadmap 04, 10
-| Capability | Repo / npm | License | Note |
-|---|---|---|---|
-| Viewer + AR launcher | google/model-viewer · `@google/model-viewer` | ✅ Apache-2.0 | `<model-viewer ar ios-src="x.usdz">`. Android WebXR/Scene Viewer handled; iOS needs a `.usdz`. |
-| USDZ on Vercel (primary) | three `USDZExporter` (`three/addons/exporters/USDZExporter.js`) | ✅ MIT | **Only zero-native-dep server path:** `await new USDZExporter().parseAsync(scene)`. |
-| USDZ high-fidelity (fallback) | google/usd_from_gltf (Docker) | ✅ Apache-2.0 (archived 2024) | Native Pixar USD → Cloud Run only, call over HTTP. Use only if fidelity demands it. |
-
-## 3. Headless WebGL / Three.js CI smoke — for roadmap 01, 04
-| Capability | Repo / npm | License | Note |
-|---|---|---|---|
-| Browser-real smoke (primary) | `@playwright/test` (already a dep) | ✅ Apache-2.0 | Real WebGL headless via ANGLE+SwiftShader, no GPU. `toHaveScreenshot()`. Bump timeouts (SwiftShader slow). |
-| Pure-Node pixel readback | stackgl/headless-gl · `gl` | ✅ BSD-2 | Feed context to `new THREE.WebGLRenderer({context})`. Native build + system GL on runner. jsdom alone has no WebGL. |
-
-## 4. Audio-driven lipsync / visemes — for roadmap 03 (note: `audio-mcp` + `a2f-nvidia` tests already exist)
-| Capability | Repo / npm | License | Note |
-|---|---|---|---|
-| Real-time browser (primary) | wass08/wawa-lipsync · `wawa-lipsync` | ✅ MIT | 100% in-browser Web Audio, zero deps. `connectAudio()`+`processAudio()` in RAF → morph targets. |
-| Real-time MFCC-accurate | mrxz/wLipSync · `wlipsync` | ✅ MIT | WASM uLipSync port; renderer-agnostic viseme weights. |
-| Turnkey talking-avatar | met4citizen/TalkingHead (+HeadTTS) | ✅ MIT | Full Oculus OVR + ARKit viseme pipeline on standard humanoid/Mixamo GLB. |
-| Precomputed | DanielSWolf/rhubarb-lip-sync · `rhubarb-lip-sync-wasm` | ✅ MIT-equiv | WASM in a Vercel fn on TTS audio → viseme JSON track for deterministic playback. |
-
-## 5. Text/image-to-3D hosted APIs — for roadmap 02, 06, 07 (TRELLIS already powers `forge_free`)
-| Model | Repo · endpoint | License | Note |
-|---|---|---|---|
-| **TRELLIS** (top pick) | microsoft/TRELLIS · Replicate `firtoz/trellis` · free NVIDIA NIM | ✅ MIT | Cleanest license; already wired. |
-| **TripoSR** (fastest) | VAST-AI/TripoSR · Replicate `camenduru/tripo-sr` | ✅ MIT | Single-image, cheap/fast. |
-| **InstantMesh** (multi-view) | TencentARC/InstantMesh · Replicate `camenduru/instantmesh` | ✅ Apache-2.0 | Image→multi-view→mesh; also covers §6. |
-| Hunyuan3D-2.1/3.1 (best texture) | Tencent-Hunyuan/Hunyuan3D-2.1 · Replicate `tencent/hunyuan-3d-3.1` | ⚠️ Tencent Community | **>1M MAU prohibited; void in EU/UK/South Korea.** Bake into terms. |
-| Stable-Fast-3D / SPAR3D | Stability-AI/stable-fast-3d | ⚠️ Stability Community | Free commercial only ≤ $1M revenue. Prefer an MIT model. |
-| License-clean self-host | TripoSG (MIT), Step1X-3D (Apache-2.0), Direct3D-S2 (MIT), Hi3DGen (MIT) | ✅ | Deploy behind Replicate-custom/Modal/RunPod. |
-
-## 6. Sketch-to-3D, multi-view, photogrammetry, splatting — for roadmap 07
-| Capability | Repo · endpoint | License | Note |
-|---|---|---|---|
-| Multi-view reconstruction | InstantMesh (see §5) | ✅ Apache-2.0 | Primary hosted multi-view path. |
-| Sketch-to-3D | HF `linoyts/sketch-to-3d` (TRELLIS) · `VAST-AI/TripoSG-scribble` | ✅ MIT | Feed clean sketch as input image to TRELLIS/TripoSR. |
-| Photogrammetry SfM | colmap/colmap | ✅ BSD-3 | GPU C++ binary; GPU-host orchestration. |
-| Splat trainer (commercial) | nerfstudio-project/gsplat | ✅ Apache-2.0 | Clean-room rasterizer; GPU training. |
-| Splat in-browser | ArthurBrussee/brush | ✅ Apache-2.0 | Rust+WebGPU+WASM; train in visitor's browser, zero server GPU. |
-| Splat viewer/editor | playcanvas/supersplat | ✅ MIT | Client-side WASM/WebGL editor, embeddable from Vercel. |
-| ⛔ AVOID | graphdeco-inria/gaussian-splatting + `diff-gaussian-rasterization` | non-commercial | License travels even on Replicate wrappers — use gsplat. |
-
-## 7. PBR material editing + AI re-texturing — for roadmap 06
-| Capability | Repo / npm | License | Note |
-|---|---|---|---|
-| Material/IBL base | `three` (`MeshPhysicalMaterial`,`PMREMGenerator`,`HDRLoader`) | ✅ MIT | Already in stack. |
-| Editor GUI | `lil-gui` (alt `tweakpane`) | ✅ MIT | Bind controllers to material props for a live PBR panel. |
-| Accurate preview | gkjohnson/three-gpu-pathtracer | ✅ MIT | Path-traced final-frame preview; lazy-load. |
-| Polish | pmndrs `postprocessing` | ✅ Zlib | Bloom/tonemapping. |
-| AI re-texture (primary) | TRELLIS via Replicate `firtoz/trellis`/fal | ✅ MIT | Only fully-permissive commercial option; gate behind $THREE/x402. |
-| ⛔ AVOID | TEXTure, Text2Tex, Paint3D, Hunyuan3D-Paint | non-commercial/capped | Research-only or territory/MAU-restricted. |
-
-## 8. Text→scene composition + editor enhancements — for roadmap 05
-| Capability | Repo / npm | License | Note |
-|---|---|---|---|
-| Scene-layout planner (primary) | weixi-feng/LayoutGPT | ✅ MIT | Re-port the LLM bbox-layout prompt into a Node/Vercel fn via the worker proxy — no Python dep. |
-| Manipulation gizmo | three `TransformControls` | ✅ MIT | Already in dep. |
-| Fast picking | `three-mesh-bvh` | ✅ MIT | Raycast/hover-select at scale. |
-| Clean imports | `three-stdlib`, `@pmndrs/vanilla` | ✅ MIT | Tree-shakeable examples + framework-free helpers. |
-| ⛔ AVOID | LayoutVLM (unlicensed), SpatialLM (CC-BY-NC) | — | Not commercially usable. |
-
-## 9. Solana SPL NFT minting + metadata + provenance — for roadmap 08 (deps already in package.json)
-| Capability | npm | License | Note |
-|---|---|---|---|
-| NFT standard (new mints) | `@metaplex-foundation/mpl-core` | ✅ Apache-2.0 | Single-account, cheaper rent, built-in royalty/freeze/provenance plugins. **Already a dep.** |
-| Legacy/pNFT compat | `@metaplex-foundation/mpl-token-metadata` | ✅ Apache-2.0 | Only for legacy SPL/pNFT. |
-| Required substrate | `@metaplex-foundation/umi` (+ bundle-defaults) | ✅ MIT | **Already a dep.** Vite: add `vite-plugin-node-polyfills` for Buffer/crypto. |
-| Raw tx / RPC | `@solana/kit` (web3.js v2 successor) | ✅ MIT | Current recommended SDK. |
-| Signing | `@noble/ed25519` | ✅ MIT | Prefer noble for new provenance signing. |
-| Permanent metadata upload | `@metaplex-foundation/umi-uploader-irys` | ✅ MIT | `uploadJson()` → Arweave URI into mpl-core create, server-side. **Bundlr is now Irys — do NOT use the old bundlr packages.** |
-
-## 10. Embeddable 3D + oEmbed + server-side OG/thumbnail — for roadmap 10
-| Capability | npm / source | License | Vercel verdict |
-|---|---|---|---|
-| Embed component | `@google/model-viewer` | ✅ Apache-2.0 | Standalone `/embed/:id` page consumers iframe. |
-| oEmbed provider | oembed.com spec (no lib) | spec | Hand-roll `/api/oembed` returning `type:"rich"` + `thumbnail_url`; add discovery `<link>`. ~30 lines. |
-| GLB→PNG (primary) | `poppygl` | ✅ MIT | **Pure-JS software rasterizer — runs on vanilla Vercel Node fn, zero native deps.** Ideal for `/api/og/:id.png`. |
-| GLB→PNG (high fidelity) | `@shopify/screenshot-glb` | ✅ MIT | Puppeteer+model-viewer. On Vercel use `puppeteer-core`+`@sparticuz/chromium`; **needs `--enable-unsafe-swiftshader --use-gl=angle --use-angle=swiftshader`** or WebGL renders blank. |
+## 1. GLB/glTF compression (Draco + meshopt), for roadmap 02, 04
+| Capability | Repo / npm | License | Status | Note |
+|---|---|---|---|---|
+| Build-time and server-side pipeline | donmccurdy/glTF-Transform · `@gltf-transform/core`+`/functions` | ✅ MIT | **[wired]** | The compression rail is `api/_lib/glb-compress.js` (`output_format: glb-draco` / `glb-meshopt`), plus roughly a dozen other `api/_lib/*` passes (bake, cleanup, themer, composer, diorama). Extend those, do not add a parallel pipeline. |
+| CLI form of the same | `@gltf-transform/cli` | ✅ MIT | **[open]** | Not a dependency. The library API already covers the server-side path, so add the CLI only for an actual build step (`gltf-transform optimize in.glb out.glb --compress meshopt`). |
+| Draco runtime decode | three `DRACOLoader` + google/draco | ✅ Apache-2.0 | **[wired]** | Decoder assets are vendored at `public/three/draco/gltf/`, not pulled from the `draco3d` npm package. `src/forge-export.js` points `setDecoderPath` there. Reuse that path rather than shipping a second copy of the wasm. |
+| Meshopt runtime decode (prefer) | zeux/meshoptimizer · `meshoptimizer` | ✅ MIT | **[wired]** | `setMeshoptDecoder(MeshoptDecoder)` across roughly nine viewer modules (`src/viewer.js`, `src/walk.js`, `src/stage.js`, others). Also the CLAUDE.md worker rule: caller-supplied glTF is meshopt-decoded through `gltf_meshopt.decode_if_meshopt` before trimesh reads it. |
+| `gltfpack` binary | zeux/meshoptimizer | ✅ MIT | **[wired]** | Not an npm dep; the pinned binary ships inside the mesh-consuming worker images. A new mesh worker inherits that requirement. |
+
+## 2. Web AR + GLB→USDZ (iOS Quick Look), for roadmap 04, 10
+| Capability | Repo / npm | License | Status | Note |
+|---|---|---|---|---|
+| Viewer + AR launcher | google/model-viewer | ✅ Apache-2.0 | **[wired]** | Loaded as a CDN script tag, **not** as an npm dependency: `@google/model-viewer` is absent from `package.json` on purpose. Used by `src/account.js` (off-screen thumbnail capture), `src/worlds-lobby.js`, `src/cosmos.js`, `src/forge-optimize.js`, and vendored at `pages/ibm/vendor/model-viewer.min.js`. Version drift is real here: live pages load 3.5.0 and 4.0.0 from three different CDNs. Match the neighboring page rather than introducing a fourth. |
+| USDZ export | three `USDZExporter` | ✅ MIT | **[wired]** | `src/usdz-pipeline.js` and `src/forge-export.js`. Note the documented limitation in `usdz-pipeline.js`: the three.js exporter does not carry skinning, so animated USDZ takes the separate path in `src/usdz-animated.js`. |
+| USDZ high-fidelity fallback | google/usd_from_gltf (Docker) | ✅ Apache-2.0 (archived 2024) | **[open]** | Native Pixar USD. On Cloud Run this is a normal sidecar service now rather than an exotic one, but the in-process exporter still covers the shipped cases. Reach for it only when fidelity genuinely demands it. |
+
+## 3. Headless WebGL / Three.js CI smoke, for roadmap 01, 04
+| Capability | Repo / npm | License | Status | Note |
+|---|---|---|---|---|
+| Browser-real smoke | `@playwright/test` | ✅ Apache-2.0 | **[wired]** | 43 specs under `tests/e2e/`. Real WebGL headless via ANGLE + SwiftShader, no GPU. Bump timeouts, SwiftShader is slow. A vitest failure gates the whole Playwright stage (`npm test`). |
+| Pure-Node pixel readback | stackgl/headless-gl · `gl` | ✅ BSD-2 | **[open]** | Not a dependency, and mostly superseded here: the server-side renderers already produce real pixels through headless chromium (`api/_lib/render-glb.js`). Consider `gl` only for a pure-Node assert that must not boot a browser. |
+
+## 4. Audio-driven lipsync / visemes, for roadmap 03
+| Capability | Repo / npm | License | Status | Note |
+|---|---|---|---|---|
+| Real-time browser | wass08/wawa-lipsync · `wawa-lipsync` | ✅ MIT | **[wired]** | Powers `/lipsync` (`public/demos/lipsync-tts.html`), which imports it as a bare specifier resolved by the vite demo-HTML rollup inputs. Its visemes are mapped onto Oculus-named morph targets in that file. |
+| In-app lipsync drivers | hand-rolled | n/a | **[wired]** | Two of them, and they are not wawa: `src/voice/lipsync-driver.js` (three-band energy estimate off a live `AnalyserNode`, for streaming TTS with no viseme timestamps) and `src/runtime/lipsync.js` (text-to-viseme heuristic, no audio analysis). Replacing either is a behavior change to a shipped surface, so flag it and prove it in a browser. |
+| Real-time MFCC-accurate | mrxz/wLipSync · `wlipsync` | ✅ MIT | **[open]** | WASM uLipSync port, renderer-agnostic viseme weights. The upgrade path if the energy-based driver proves too coarse. |
+| Turnkey talking-avatar | met4citizen/TalkingHead (+HeadTTS) | ✅ MIT | **[open]** | Full Oculus OVR + ARKit viseme pipeline on standard humanoid/Mixamo GLB. |
+| Precomputed | DanielSWolf/rhubarb-lip-sync · `rhubarb-lip-sync-wasm` | ✅ MIT-equiv | **[open]** | WASM pass over TTS audio producing a deterministic viseme JSON track. |
+| Adjacent, check first | `packages/audio-mcp/` | n/a | **[wired]** | TTS, STT, audio-to-face lipsync and motion capture already live here. Read it before adding any lipsync dependency. |
+
+## 5. Text/image-to-3D hosted APIs, for roadmap 02, 06, 07
+| Model | Repo · endpoint | License | Status | Note |
+|---|---|---|---|---|
+| **TRELLIS** (top pick) | microsoft/TRELLIS · Replicate `firtoz/trellis` · free NVIDIA NIM | ✅ MIT | **[wired]** | Powers the free forge lane. Cleanest license of the set. |
+| **TripoSR** (fastest) | VAST-AI/TripoSR · Replicate `camenduru/tripo-sr` | ✅ MIT | **[open]** | Single-image, cheap and fast. |
+| **InstantMesh** (multi-view) | TencentARC/InstantMesh · Replicate `camenduru/instantmesh` | ✅ Apache-2.0 | **[open]** | Image to multi-view to mesh; also covers §6. |
+| Hunyuan3D-2.1/3.1 (best texture) | Tencent-Hunyuan/Hunyuan3D-2.1 | ⚠️ Tencent Community | **[open]** | **>1M MAU prohibited; void in EU/UK/South Korea.** Bake into terms before shipping, not after. |
+| Stable-Fast-3D / SPAR3D | Stability-AI/stable-fast-3d | ⚠️ Stability Community | **[open]** | Free commercial only up to $1M revenue. Prefer an MIT model. |
+| License-clean self-host | TripoSG (MIT), Step1X-3D (Apache-2.0), Direct3D-S2 (MIT), Hi3DGen (MIT) | ✅ | **[open]** | Deploy behind a `workers/<name>` Cloud Run GPU service, Replicate-custom, Modal, or RunPod. GCP is the pre-approved surface (CLAUDE.md standing resource approvals). |
+
+## 6. Sketch-to-3D, multi-view, photogrammetry, splatting, for roadmap 07
+| Capability | Repo · endpoint | License | Status | Note |
+|---|---|---|---|---|
+| Multi-view reconstruction | InstantMesh (see §5) | ✅ Apache-2.0 | **[open]** | Primary hosted multi-view path. |
+| Sketch-to-3D | HF `linoyts/sketch-to-3d` (TRELLIS) · `VAST-AI/TripoSG-scribble` | ✅ MIT | **[open]** | Feed a clean sketch as the input image to TRELLIS/TripoSR. |
+| Photogrammetry SfM | colmap/colmap | ✅ BSD-3 | **[open]** | GPU C++ binary, so a GPU worker service. |
+| Splat trainer (commercial) | nerfstudio-project/gsplat | ✅ Apache-2.0 | **[open]** | Clean-room rasterizer; GPU training. |
+| Splat in-browser | ArthurBrussee/brush | ✅ Apache-2.0 | **[open]** | Rust + WebGPU + WASM; trains in the visitor's browser, zero server GPU. |
+| Splat viewer/editor | playcanvas/supersplat | ✅ MIT | **[open]** | Client-side WASM/WebGL editor, embeddable. |
+| ⛔ AVOID | graphdeco-inria/gaussian-splatting + `diff-gaussian-rasterization` | non-commercial | | The license travels even through Replicate wrappers. Use gsplat. |
+
+## 7. PBR material editing + AI re-texturing, for roadmap 06
+| Capability | Repo / npm | License | Status | Note |
+|---|---|---|---|---|
+| Material/IBL base | `three` (`MeshPhysicalMaterial`, `PMREMGenerator`, `HDRLoader`) | ✅ MIT | **[wired]** | Already the stack. |
+| Polish | pmndrs `postprocessing` | ✅ Zlib | **[wired]** | Roughly ten call sites. Bloom, tonemapping. |
+| Accurate preview | gkjohnson/three-gpu-pathtracer | ✅ MIT | **[wired]** | One call site. Path-traced final-frame preview; keep it lazy-loaded. |
+| Editor GUI | `lil-gui` | ✅ MIT | **[dep-only]** | Declared in `package.json` at `^0.21.0` and imported by nothing in the repo. Either use it for the PBR panel this section anticipated, or drop the dependency. Do not add `tweakpane` alongside it. |
+| AI re-texture | TRELLIS via Replicate/fal | ✅ MIT | **[open]** | The only fully permissive commercial option here. Gate behind $THREE / x402. |
+| ⛔ AVOID | TEXTure, Text2Tex, Paint3D, Hunyuan3D-Paint | non-commercial/capped | | Research-only or territory/MAU-restricted. |
+
+## 8. Text to scene composition + editor enhancements, for roadmap 05
+| Capability | Repo / npm | License | Status | Note |
+|---|---|---|---|---|
+| Scene-layout planner | weixi-feng/LayoutGPT | ✅ MIT | **[open]** | Re-port the LLM bbox-layout prompt into a Node handler via the worker proxy, no Python dependency. |
+| Manipulation gizmo | three `TransformControls` | ✅ MIT | **[wired]** | Ships with three. |
+| Fast picking | `three-mesh-bvh` | ✅ MIT | **[wired]** | Two call sites. Raycast/hover-select at scale. |
+| Clean imports | `three-stdlib`, `@pmndrs/vanilla` | ✅ MIT | **[open]** | Neither is a dependency. The repo imports `three/addons/*` directly, which is the established pattern; match it rather than introducing a second convention. |
+| ⛔ AVOID | LayoutVLM (unlicensed), SpatialLM (CC-BY-NC) | | | Not commercially usable. |
+
+## 9. Solana SPL NFT minting + metadata + provenance, for roadmap 08
+| Capability | npm | License | Status | Note |
+|---|---|---|---|---|
+| NFT standard (new mints) | `@metaplex-foundation/mpl-core` | ✅ Apache-2.0 | **[wired]** | `api/nft/mint-scene.js`, `api/_lib/skill-nft.js`, `api/_lib/draft-mint.js`, `api/_lib/tokenize-3d.js` and more. Single-account, cheaper rent, built-in royalty/freeze/provenance plugins. |
+| Required substrate | `@metaplex-foundation/umi` (+ bundle-defaults) | ✅ MIT | **[wired]** | Note `api/nft/mint-scene.js`: umi is constructed over the failover `Connection`, so a single node's malformed 200 fails over instead of throwing a StructError. Preserve that when you add a mint path. |
+| Raw tx / RPC | `@solana/kit` | ✅ MIT | **[wired]** | Present alongside `@solana/web3.js`. |
+| Signing | `@noble/curves/ed25519` | ✅ MIT | **[wired]** | **Correction to the June survey, which recommended the standalone `@noble/ed25519`.** The repo standardized on the `@noble/curves` bundle (15 modules import `@noble/curves/ed25519`, zero import `@noble/ed25519`), alongside `secp256k1`, `bls12-381` and `@noble/hashes`. Adding the standalone package would duplicate a dependency already present. |
+| Metadata permanence | `@metaplex-foundation/umi-uploader-irys` | ✅ MIT | **[open]** | **Not the road taken.** Mint metadata is pinned to IPFS through `pinAsset` (`api/_lib/ipfs-gateways.js`) and the resulting URI goes into `createV1`. Adding Irys/Arweave would be a second permanence rail, not a drop-in. Decide deliberately. (Bundlr is now Irys; the old bundlr packages are dead either way.) |
+| Legacy/pNFT compat | `@metaplex-foundation/mpl-token-metadata` | ✅ Apache-2.0 | **[open]** | Not a dependency. Only needed for legacy SPL/pNFT. |
+
+## 10. Embeddable 3D + oEmbed + server-side OG/thumbnail, for roadmap 10
+| Capability | npm / source | License | Status | Note |
+|---|---|---|---|---|
+| Embed component | model-viewer | ✅ Apache-2.0 | **[wired]** | See §2 on the CDN-not-npm reality. |
+| oEmbed provider | oembed.com spec (no library) | spec | **[wired]** | Hand-rolled exactly as this file predicted: `api/agent-oembed.js` (routed from `/api/oembed`) and `api/play-oembed.js`, plus `/api/widgets/oembed`. Extend those. |
+| GLB→PNG | `puppeteer-core` + `@sparticuz/chromium-min` | ✅ Apache-2.0 | **[wired]** | **Correction to the June survey**, which named `poppygl` primary and `@shopify/screenshot-glb` (plain `@sparticuz/chromium`) the fallback. Neither was adopted. Three renderers share the headless-chromium approach: `api/_lib/render-glb.js` (OG cards), `api/_lib/avatar-render.js` (the public render endpoint and the `render_avatar_image` MCP tool) and `api/_lib/render-clip.js` (`render_avatar_clip`). The `-min` build downloads its chromium pack at first use, pinned in lockstep with the package version. |
+| The renderers' three.js supply | `api/_lib/three-cdn.js` | n/a | **[wired]** | All three renderers build a page whose import map points at a pinned three.js release, and they resolve the host through `resolveThreeCdn()` (unpkg, falling back to jsDelivr on a bounded HEAD probe, cached ten minutes). `THREE_VERSION` lives in that one module. Do not hardcode a CDN host in a renderer; `tests/api/three-cdn-wiring.test.js` fails the build if you do. |
+| GLB→PNG, pure JS | `poppygl` | ✅ MIT | **[open]** | Software rasterizer, zero native deps. Its original selling point was surviving a Vercel function, which no longer constrains us. Still the cheapest option if a lane ever needs a poster without paying chromium's cold start. |
---
## Top-line picks
-- **Compression:** `@gltf-transform/cli` (build) + `MeshoptDecoder` (runtime); Draco for max-compression assets.
-- **AR:** `<model-viewer>` + three `USDZExporter` on Vercel; Cloud Run `usd_from_gltf` only if fidelity demands.
-- **CI:** Playwright (already a dep) for canvas smoke; `gl` for pure-Node pixel asserts.
-- **Lipsync:** wawa-lipsync (live) + Rhubarb-WASM (precomputed) — but check `audio-mcp` first, it may already cover this.
-- **Gen-3D:** TRELLIS (MIT, already wired), TripoSR (fast), InstantMesh (multi-view); Hunyuan only with MAU/EU-UK-KR terms baked in.
-- **Splatting:** gsplat/Brush + SuperSplat — never the Inria 3DGS rasterizer.
-- **Solana:** mpl-core + Umi + umi-uploader-irys + @solana/kit + @noble/ed25519 (mpl-core, umi already deps).
-- **Embed/OG:** `<model-viewer>` + hand-rolled oEmbed + poppygl (screenshot-glb as fidelity fallback).
-
-**Hard AVOID:** Inria 3D Gaussian Splatting, TEXTure/Text2Tex/Paint3D, LayoutVLM, SpatialLM weights, Meshroom (serverless), Blender headless. **Read fine print (⚠️):** Hunyuan3D (1M MAU + EU/UK/KR exclusion), SF3D/SPAR3D ($1M revenue cap).
+
+- **Compression:** the `@gltf-transform` passes in `api/_lib/` (server) plus `MeshoptDecoder` (runtime); vendored Draco decoder for max-compression assets.
+- **AR:** model-viewer plus three's `USDZExporter`; a `usd_from_gltf` sidecar only if fidelity demands it.
+- **CI:** Playwright for canvas smoke; `gl` only for a pure-Node pixel assert.
+- **Lipsync:** wawa-lipsync is already live on `/lipsync`; read `packages/audio-mcp/` before adding anything else.
+- **Gen-3D:** TRELLIS (MIT, already wired), TripoSR (fast), InstantMesh (multi-view); Hunyuan only with MAU and EU/UK/KR terms baked in.
+- **Splatting:** gsplat/Brush plus SuperSplat, never the Inria 3DGS rasterizer.
+- **Solana:** mpl-core + umi + `@solana/kit` + `@noble/curves/ed25519`, with metadata pinned to IPFS.
+- **Embed/OG:** model-viewer, the hand-rolled oEmbed handlers, and the shared chromium renderers.
+
+**Hard AVOID:** Inria 3D Gaussian Splatting, TEXTure/Text2Tex/Paint3D, LayoutVLM, SpatialLM weights, Meshroom, headless Blender. **Read the fine print (⚠️):** Hunyuan3D (1M MAU plus EU/UK/KR exclusion), SF3D/SPAR3D ($1M revenue cap).
+
+## Open items this audit surfaced
+
+1. `lil-gui` is a dependency nobody imports (§7). Use it or drop it.
+2. model-viewer is loaded at two major versions from three CDNs across live pages (§2). Worth converging.
+3. The headless renderers pin three.js at 0.176.0 while the app runs 0.184.0. That is deliberate for now (a bump changes every OG card and needs visual verification), but the pin is a single constant in `api/_lib/three-cdn.js` when someone takes it on.
+
+## How this file was verified
+
+The status column is not a memory of what was planned. Every **[wired]**,
+**[dep-only]** and **[open]** marker above was re-derived on 2026-09-03 by
+reading `package.json` and grepping the working tree for the actual import
+sites. Re-run that when you touch this file; a status marker that has quietly
+gone stale is worse than no marker, because agents execute this file verbatim.
## Retire this file when the campaign is done (required)
diff --git a/prompts/roadmap/_generated/generation-suite/gate-before.txt b/prompts/roadmap/_generated/generation-suite/gate-before.txt
new file mode 100644
index 000000000..e9d188a0c
--- /dev/null
+++ b/prompts/roadmap/_generated/generation-suite/gate-before.txt
@@ -0,0 +1,85 @@
+
+> npm run check:conflicts && npm run check:claude && npm run check:vendored && npm run check:fetch-timeouts && npm run check:tdz-bootstrap && npm run check:event && npm run check:cron-syntax && npm run check:studio-openapi && npm run test:gate && npm run test:gate-3d && npm run audit:mcp && npm run audit:mcp-golden && npm run audit:mcp-safety && npm run audit:mcp-catalog && npm run audit:routes && npm run audit:handlers && npm run audit:pages && npm run audit:links && npm run audit:docs && npm run audit:upstreams && npm run check:tutorials && npm run audit:hidden-guard && npm run audit:inline-handlers && npm run audit:x402-catalog && npm run audit:tokens && npm run audit:guards && npm run audit:tour-atlas && npm run check:ios-icons
+
+
+> [email protected] check:conflicts
+> node scripts/check-merge-conflicts.mjs
+
+check:conflicts — no unresolved merge-conflict markers in tracked files
+
+> [email protected] check:claude
+> node scripts/check-claude-md.mjs
+
+[check-claude] OK: every script, path, and typography rule in CLAUDE.md matches the repo
+
+> [email protected] check:vendored
+> node scripts/check-vendored-workers.mjs
+
+[check-vendored-workers] OK: 27 vendored file copies across 32 workers are byte-identical
+
+> [email protected] check:fetch-timeouts
+> node scripts/check-fetch-timeouts.mjs
+
+check:fetch-timeouts: every external fetch, and every fetch under api/, is bounded by a deadline
+
+> [email protected] check:tdz-bootstrap
+> node scripts/check-tdz-bootstrap.mjs
+
+check:tdz-bootstrap: no bootstrap-before-state ordering in 1595 browser module(s)
+
+> [email protected] check:event
+> node scripts/check-event-window.mjs
+
+[check-event] OK: public/event.json carries the explicit no-event state, so no event is configured (a supported state)
+
+> [email protected] check:cron-syntax
+> node scripts/check-cron-drift.mjs --offline
+
+Declared crons in vercel.json: 114
+Most frequent:
+ every 1m /api/cron/economy-tick
+ every 1m /api/cron/three-market-refresh
+ every 1m /api/cron/avaturn-seed-cron
+ every 1m /api/cron/flush-usage-events
+ every 1m /api/cron/forge-finalize
+
+All cron expressions are valid (offline mode: live jobs not compared).
+
+> [email protected] check:studio-openapi
+> node scripts/sync-studio-openapi.mjs --check
+
+3D Studio OpenAPI in sync: prompts/store-submissions/_generated/openai-actions.yaml matches public/.well-known/3d-studio-openapi.yaml.
+
+> [email protected] test:gate
+> node scripts/test-gate.mjs
+
+[test-gate] running 7 critical-path test files…
+
+ RUN v4.1.9 /workspaces/three.ws
+
+ ❯ tests/api/healthz.test.js (0 test)
+
+⎯⎯⎯⎯⎯⎯ Failed Suites 1 ⎯⎯⎯⎯⎯⎯⎯
+
+ FAIL tests/api/healthz.test.js [ tests/api/healthz.test.js ]
+Error: no database in test
+ ❯ Proxy.sql tests/api/healthz.test.js:15:21
+ 13| // test non-deterministically.
+ 14| vi.mock('../../api/_lib/db.js', () => ({
+ 15| sql: () => { throw new Error('no database in test'); },
+ | ^
+ 16| isDbUnavailableError: () => false,
+ 17| isDbCapacityError: () => false,
+ ❯ api/_lib/home/store.js:49:25
+ ❯ api/_lib/home/runtime.js:31:1
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/1]⎯
+
+
+ Test Files 1 failed | 6 passed (7)
+ Tests 73 passed (73)
+ Start at 07:14:56
+ Duration 15.36s (transform 3.45s, setup 0ms, import 36.85s, tests 2.61s, environment 1ms)
+
+[test-gate] ✗ critical-path tests failed — blocking deploy
diff --git a/prompts/roadmap/_generated/reuse-map-audit/gate-after.txt b/prompts/roadmap/_generated/reuse-map-audit/gate-after.txt
new file mode 100644
index 000000000..2d7e2b722
--- /dev/null
+++ b/prompts/roadmap/_generated/reuse-map-audit/gate-after.txt
@@ -0,0 +1,112 @@
+
+> npm run check:conflicts && npm run check:claude && npm run check:vendored && npm run check:fetch-timeouts && npm run check:tdz-bootstrap && npm run check:event && npm run check:cron-syntax && npm run check:studio-openapi && npm run test:gate && npm run test:gate-3d && npm run audit:mcp && npm run audit:mcp-golden && npm run audit:mcp-safety && npm run audit:mcp-catalog && npm run audit:routes && npm run audit:handlers && npm run audit:pages && npm run audit:links && npm run audit:docs && npm run audit:upstreams && npm run check:tutorials && npm run audit:hidden-guard && npm run audit:inline-handlers && npm run audit:x402-catalog && npm run audit:tokens && npm run audit:guards && npm run audit:tour-atlas && npm run check:ios-icons
+
+
+> [email protected] check:conflicts
+> node scripts/check-merge-conflicts.mjs
+
+check:conflicts — no unresolved merge-conflict markers in tracked files
+
+> [email protected] check:claude
+> node scripts/check-claude-md.mjs
+
+[check-claude] OK: every script, path, and typography rule in CLAUDE.md matches the repo
+
+> [email protected] check:vendored
+> node scripts/check-vendored-workers.mjs
+
+[check-vendored-workers] OK: 27 vendored file copies across 32 workers are byte-identical
+
+> [email protected] check:fetch-timeouts
+> node scripts/check-fetch-timeouts.mjs
+
+check:fetch-timeouts: every external fetch, and every fetch under api/, is bounded by a deadline
+
+> [email protected] check:tdz-bootstrap
+> node scripts/check-tdz-bootstrap.mjs
+
+check:tdz-bootstrap: no bootstrap-before-state ordering in 1597 browser module(s)
+
+> [email protected] check:event
+> node scripts/check-event-window.mjs
+
+[check-event] OK: public/event.json carries the explicit no-event state, so no event is configured (a supported state)
+
+> [email protected] check:cron-syntax
+> node scripts/check-cron-drift.mjs --offline
+
+Declared crons in vercel.json: 114
+Most frequent:
+ every 1m /api/cron/economy-tick
+ every 1m /api/cron/three-market-refresh
+ every 1m /api/cron/avaturn-seed-cron
+ every 1m /api/cron/flush-usage-events
+ every 1m /api/cron/forge-finalize
+
+All cron expressions are valid (offline mode: live jobs not compared).
+
+> [email protected] check:studio-openapi
+> node scripts/sync-studio-openapi.mjs --check
+
+3D Studio OpenAPI in sync: prompts/store-submissions/_generated/openai-actions.yaml matches public/.well-known/3d-studio-openapi.yaml.
+
+> [email protected] test:gate
+> node scripts/test-gate.mjs
+
+[test-gate] running 7 critical-path test files…
+
+ RUN v4.1.9 /workspaces/three.ws
+
+
+ Test Files 7 passed (7)
+ Tests 86 passed (86)
+ Start at 07:25:43
+ Duration 2.93s (transform 2.66s, setup 0ms, import 5.30s, tests 2.79s, environment 1ms)
+
+[test-gate] ✓ all critical-path tests passed
+
+> [email protected] test:gate-3d
+> vitest run tests/glb-canonicalize.test.js tests/animation-retarget.test.js tests/animation-retarget-rest-pollution.test.js tests/viewer-framing.test.js
+
+
+ RUN v4.1.9 /workspaces/three.ws
+
+
+ Test Files 4 passed (4)
+ Tests 559 passed (559)
+ Start at 07:25:46
+ Duration 692ms (transform 400ms, setup 0ms, import 948ms, tests 502ms, environment 1ms)
+
+
+> [email protected] audit:mcp
+> node scripts/audit-mcp-manifests.mjs
+
+[audit:mcp] 49 MCP manifests consistent
+
+> [email protected] audit:mcp-golden
+> node scripts/audit-mcp-golden.mjs
+
+[audit:mcp-golden] api/_mcp3d/tools/sim-readiness.js: tool ADDED: grade_sim_readiness
+
+[audit:mcp-golden] 1 contract change(s) vs golden fixture.
+ If every change above is INTENTIONAL (reviewed as a public-contract change),
+ refresh the fixture: node scripts/audit-mcp-golden.mjs --update
+exit=1
+
+--- audit note (agent, 2026-09-03) ---
+gate-before failed at test:gate (tests/api/healthz.test.js could not import:
+the db mock threw synchronously, killing api/_lib/home/store.js's module-scope
+sql fragment). That failure short-circuited the chain, so gate-before never
+reached audit:mcp-golden.
+
+gate-after clears every stage that gate-before reached, plus the ones it could
+not: test:gate 7/7 (was 6 passed, 1 failed), test:gate-3d 4/4, audit:mcp 49
+manifests consistent.
+
+The single remaining failure, audit:mcp-golden reporting grade_sim_readiness
+ADDED, comes from api/_mcp3d/tools/sim-readiness.js, an UNTRACKED file belonging
+to a concurrent agent in this shared worktree. It is not part of this work and
+the golden fixture was deliberately left alone: refreshing it would rubber-stamp
+another agent's unreviewed public MCP contract change. That agent should run
+`node scripts/audit-mcp-golden.mjs --update` when it commits.
diff --git a/prompts/roadmap/_generated/reuse-map-audit/gate-before.txt b/prompts/roadmap/_generated/reuse-map-audit/gate-before.txt
new file mode 100644
index 000000000..c829f4523
--- /dev/null
+++ b/prompts/roadmap/_generated/reuse-map-audit/gate-before.txt
@@ -0,0 +1,86 @@
+
+> npm run check:conflicts && npm run check:claude && npm run check:vendored && npm run check:fetch-timeouts && npm run check:tdz-bootstrap && npm run check:event && npm run check:cron-syntax && npm run check:studio-openapi && npm run test:gate && npm run test:gate-3d && npm run audit:mcp && npm run audit:mcp-golden && npm run audit:mcp-safety && npm run audit:mcp-catalog && npm run audit:routes && npm run audit:handlers && npm run audit:pages && npm run audit:links && npm run audit:docs && npm run audit:upstreams && npm run check:tutorials && npm run audit:hidden-guard && npm run audit:inline-handlers && npm run audit:x402-catalog && npm run audit:tokens && npm run audit:guards && npm run audit:tour-atlas && npm run check:ios-icons
+
+
+> [email protected] check:conflicts
+> node scripts/check-merge-conflicts.mjs
+
+check:conflicts — no unresolved merge-conflict markers in tracked files
+
+> [email protected] check:claude
+> node scripts/check-claude-md.mjs
+
+[check-claude] OK: every script, path, and typography rule in CLAUDE.md matches the repo
+
+> [email protected] check:vendored
+> node scripts/check-vendored-workers.mjs
+
+[check-vendored-workers] OK: 27 vendored file copies across 32 workers are byte-identical
+
+> [email protected] check:fetch-timeouts
+> node scripts/check-fetch-timeouts.mjs
+
+check:fetch-timeouts: every external fetch, and every fetch under api/, is bounded by a deadline
+
+> [email protected] check:tdz-bootstrap
+> node scripts/check-tdz-bootstrap.mjs
+
+check:tdz-bootstrap: no bootstrap-before-state ordering in 1595 browser module(s)
+
+> [email protected] check:event
+> node scripts/check-event-window.mjs
+
+[check-event] OK: public/event.json carries the explicit no-event state, so no event is configured (a supported state)
+
+> [email protected] check:cron-syntax
+> node scripts/check-cron-drift.mjs --offline
+
+Declared crons in vercel.json: 114
+Most frequent:
+ every 1m /api/cron/economy-tick
+ every 1m /api/cron/three-market-refresh
+ every 1m /api/cron/avaturn-seed-cron
+ every 1m /api/cron/flush-usage-events
+ every 1m /api/cron/forge-finalize
+
+All cron expressions are valid (offline mode: live jobs not compared).
+
+> [email protected] check:studio-openapi
+> node scripts/sync-studio-openapi.mjs --check
+
+3D Studio OpenAPI in sync: prompts/store-submissions/_generated/openai-actions.yaml matches public/.well-known/3d-studio-openapi.yaml.
+
+> [email protected] test:gate
+> node scripts/test-gate.mjs
+
+[test-gate] running 7 critical-path test files…
+
+ RUN v4.1.9 /workspaces/three.ws
+
+ ❯ tests/api/healthz.test.js (0 test)
+
+⎯⎯⎯⎯⎯⎯ Failed Suites 1 ⎯⎯⎯⎯⎯⎯⎯
+
+ FAIL tests/api/healthz.test.js [ tests/api/healthz.test.js ]
+Error: no database in test
+ ❯ Proxy.sql tests/api/healthz.test.js:15:21
+ 13| // test non-deterministically.
+ 14| vi.mock('../../api/_lib/db.js', () => ({
+ 15| sql: () => { throw new Error('no database in test'); },
+ | ^
+ 16| isDbUnavailableError: () => false,
+ 17| isDbCapacityError: () => false,
+ ❯ api/_lib/home/store.js:49:25
+ ❯ api/_lib/home/runtime.js:31:1
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/1]⎯
+
+
+ Test Files 1 failed | 6 passed (7)
+ Tests 73 passed (73)
+ Start at 07:19:04
+ Duration 2.72s (transform 2.06s, setup 0ms, import 3.18s, tests 2.77s, environment 1ms)
+
+[test-gate] ✗ critical-path tests failed — blocking deploy
+exit=1