Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
JavaScript 3d-generationai-agentsanimationavatardigital-humangenerative-aiglbgltfimage-to-3dllmmcpmcp-servermodel-context-protocolsolanatext-to-3dthreejsweb-componentswebglwebxrx402
docs(home): record the order 18 re-verification and the three doc claims it disproved
nirholas60441cc6f32b
diff --git a/prompts/finish/_context/home-PROGRESS.md b/prompts/finish/_context/home-PROGRESS.mdindex f60f99409..79f29613d 100644--- a/prompts/finish/_context/home-PROGRESS.md+++ b/prompts/finish/_context/home-PROGRESS.md@@ -935,3 +935,85 @@ byte-identical to one the surviving migration creates and each was verified to bbefore the drop; after applying, zero orphans remain, all four intended indexes are present, thegrants index is still UNIQUE and the row counts are unchanged.++---++## 18 (re-verified). Docs, SDK publish, the home MCP server package (2026-09-09)++**Why a second pass:** the order file's own step 0 says nothing in it is a status claim to trust,+so every line of the 2026-09-03 entry above was re-measured against the tree six days later. Three+things had drifted, all of them the kind that reads as true until someone runs it.++**Fixed:**++1. **`packages/home-mcp`'s README documented test commands that cannot work.** The suite moved to+ vitest at `packages/home-mcp/tests/*.test.js`, and the README still said+ `node --test "packages/home-mcp/test/**/*.test.mjs"` (singular `test/`, `.test.mjs`). The glob+ form matched nothing and **exited 0**, so a reader would have concluded the package was green+ without running one assertion; the single-file form failed with `Could not find`. Both replaced+ with the vitest commands, both executed.+2. **The live gate suite was flaky by construction and was failing.** `pickLastSettledLock` chose+ the last lock sitting at `locked` or `unlocked`, with a comment claiming that filter kept it off+ the demo integration's deliberately unreliable lock. It does not: `lock.poorly_installed_door`+ sits at `locked` on a fresh instance and only jams once something acts on it, so on a house+ nobody had touched yet it was picked, sorted last, and then `beforeAll` timed out waiting for it+ to lock. It passed on 2026-09-03 only because an earlier suite had already jammed it on the+ shared instance. Replaced with `claimLockableLock`, which walks candidates from the far end and+ accepts one only after it has actually reached `locked`, so the resting state is never taken as+ evidence.+3. **`docs/smart-home.md` listed a shipped feature under "Not shipped".** Its bullet said a+ floorplan editor "is not built" while `src/home/floorplan.js` (774 lines), `api/_lib/home/layout.js`,+ `api/home/[id]/layout.js` and two migrations were in the tree and mounted from `src/home/scene.js`.+ Moved into the shipped table; the not-shipped bullet now names what is genuinely absent (room+ shapes other than a rectangle: rotation, wall openings and polygons, which exist only as reserved+ fields behind `LAYOUT_FORMAT`). The same section's migration count said 8; `api/_lib/migrations/*_home_*.sql`+ is 11. `STRUCTURE.md`'s 3D-home row named the scene's model, renderer, fallback and controller but+ none of the floorplan files, so the editor was unfindable from the map; added.++**Measured** (house: `node scripts/home-test-instance.mjs --up --onboard --seed --name docs18`,+Home Assistant 2026.9.0, 1 floor, 4 areas, 4 locks, 2 scenes, port 42785):++- `npx vitest run packages/home-mcp`: 21/21 passing with the house configured, 13 passing and 8+ live tests correctly skipped without it. Includes "refuses a guarded unlock, and the door does+ not move" and "lets the operator's own out-of-band allowance through, and it really unlocks".+- `npx vitest run packages/home-bridge`: 47/47.+- Both READMEs' code examples executed against that house. `home-bridge`: connect and room graph,+ a real `light.turn_on`, `activate('good night')` resolving to `scene.bedtime` at 0.95, the gate+ raising `needs_confirmation` then running under `{confirmed:true}`, the per-entity allow list,+ the MCP channel (29 real tools plus a tool call that moved a real light), and the relay+ transport's URL shape. `home-mcp`: the stdio client printed the real rooms, `refused: true`, and+ `lock.front_door` read `locked` afterwards.+- Tutorial, step by step: `curl` reachability `200`; `/api/` returning `{"message":"API running."}`+ and `401` on a bad token; the harness one-liner; the stderr banner, verbatim as documented+ (`[[email protected]] connected over stdio with 5 tools, home http://127.0.0.1:42785`); section 4's+ three JSON payloads, which match what the tools really return, including the no-match refusal+ and the `goodnight` / `bedtime` / `time for bed` synonyms all landing on `scene.bedtime`;+ section 8's error contract, where `assertDialableHomeUrl` raises `private_address` internally and+ `api/home/index.js` maps it to the documented wire code `unreachable` with the message verbatim.+- The tutorial's load-bearing security claim re-proved live: `intent__HassTurnOff` is described by+ Home Assistant itself as *"Turns off/closes a device or entity. For locks, this performs an+ 'unlock' action."* and calling it on a `locked` front door left it `unlocked`.+- `npm pack --dry-run`: `@three-ws/home-bridge` 12 files / 33.8 kB, `@three-ws/home-mcp` 12 files /+ 18.3 kB. No tests, no fixtures, no tokens in either; `home-mcp`'s `files` matches `brain-mcp`'s+ convention exactly, so `smithery.yaml` stays out of the tarball on purpose.+- `npm run audit:docs`: clean, 1586 files. `npm run check:claude`: OK. `npm run check:rules`: clean.+ `npm run check:docs-search`: current after a rebuild (690 docs, 8976 sections); it is gitignored+ and goes stale again within minutes here because peers are editing docs continuously, so its+ staleness is a shared-worktree artifact and never something to commit.+- `npm run publish:packages:dry` and `npm run publish:mcp:dry`: both clean, both packages still+ reported as would-publish.++**Deviations from the order file:** the file's docs table asks for `/home` and `/home/:id` in+`data/pages.json`. Those routes do not exist; `/home` is a 301 to `/`. The real ones are+`/smart-home` and `/smart-home/:id`, and `data/pages.json` already carries `/smart-home` and its+four static children plus every home doc and the tutorial. The 2026-09-03 entry recorded the same+correction; it is repeated here because the order file still says otherwise and was left on disk.++**Left open:** the npm publish, still owner-gated and still the only outstanding step, so the order+file stays on disk per its own retire clause. No changelog entry was added: this pass corrected+documentation of features whose user-visible shipping already has entries (the floorplan editor's+landed the same day), and a "we fixed our own doc" line would be noise in a community feed.++**Commits:** `30972fa08` (the vitest command and the gate-test pick, swept into a peer's commit+before I could stage it, under an accurate message), `54b71805a` (the `docs/smart-home.md`+corrections, likewise swept), `934ae7aea` (the STRUCTURE.md row).