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(best3d): log order 070 close-out, add owner-deploy rows for both workers

nirholascf6f189fa530
diff --git a/prompts/finish/070-best3d-04-editable-parts.md b/prompts/finish/070-best3d-04-editable-parts.md
index ae98179f3..95dd00702 100644
--- a/prompts/finish/070-best3d-04-editable-parts.md
+++ b/prompts/finish/070-best3d-04-editable-parts.md
@@ -1,5 +1,7 @@
# best3d 04: Editable parts: PartCrafter and HoloPart behind the segment lane
+**Evolve status (2026-10-11):** Built and tested, NOT closed. `workers/model-partcrafter/` (image-to-parts) and `workers/model-holopart/` (occluded-part completion) are both complete, with `api/_lib/forge-tiers.js`/`api/_lib/forge-options.js`/`api/_providers/gcp.js`/`api/forge.js`/`api/gpt-forge.js`/`api/forge-holopart.js` wiring both backends end to end, opt-in only (never a default or failover target). Restyle Studio (`src/restyle.js`/`pages/restyle.html`) gained a Parts panel: select a part, recolor/retexture it alone via `editRoot()` scoping, swap it from a file, or regenerate it through `/api/forge-holopart`, then reassemble under the original node name (satisfies task 3, order 078's dependency). DoD line 1 ("a real image yields at least 3 named part nodes and one part is retextured without touching the others") is verified against a synthetic 3-part GLB fixture in a real browser (Playwright, zero console errors), matching the exact scene-graph shape `workers/segment/segment_core.py` already produces in production, but NOT yet against a real image run through a live PartCrafter/HoloPart service, since neither is deployed. DoD lines 2 and 3 (tests, docs/changelog) are fully done: `npm test` passes for every touched file, `npm run audit:docs` reports nothing new, `STRUCTURE.md`/`docs/forge.md`/`data/changelog.json` all updated and `npm run build:pages` run. This pass also found and fixed a real pre-existing gap discovered while verifying this order: `workers/model-holopart/Dockerfile` (and, separately, `workers/rig-nonhuman/Dockerfile`) never staged the `gltfpack` binary their vendored `gltf_meshopt.py` decode step depends on at runtime; both now do, and `scripts/check-vendored-workers.mjs` gained a mechanical check so this class of gap cannot regress silently. Remaining, owner-gated: deploy `model-partcrafter` and `model-holopart` to Cloud Run and wire their URLs (rows 43 and 44 of [production-100-OWNER-ACTIONS.md](_context/production-100-OWNER-ACTIONS.md)), then run one real image/mesh through each and confirm DoD line 1 against live output before deleting this file.
+
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)
diff --git a/prompts/finish/_context/best3d-PROGRESS.md b/prompts/finish/_context/best3d-PROGRESS.md
index 3e1360d80..fd62fe5bb 100644
--- a/prompts/finish/_context/best3d-PROGRESS.md
+++ b/prompts/finish/_context/best3d-PROGRESS.md
@@ -21,3 +21,8 @@ Append a dated entry per closed order, with commit SHAs.
- Still open, owner (rows 40 and 41 of `_context/production-100-OWNER-ACTIONS.md`): (1) `gcloud builds submit --config workers/rig-nonhuman/cloudbuild.yaml` and set `RIG_NONHUMAN_URL`; (2) `gcloud builds submit --config workers/model-anigen/cloudbuild.yaml` and set `MODEL_ANIGEN_URL` on `three-ws-api`; (3) once both are live, run one real quadruped GLB through each lane, confirm it plays a clip in the viewer, then delete the order file.
- Commits: `ac2eca062` (non-humanoid rig worker + dispatch), `ba76f8173` (AniGen worker), `7f2646b67` (AniGen Forge wiring), `af041fb9d` (docs: STRUCTURE.md + changelog for both), `fe5c55f0c` (test: glb-canonicalize non-humanoid coverage), `a7906b7c9` (feat: Rig Doctor skin-weight report), `aa3b812da` (docs: changelog for the Rig Doctor stat), `f10073da8` (docs: owner-action rows 40/41).
- 2026-10-11 (retry pass, same day): re-ran order 069 end to end, nothing drifted and no code changed. Confirmed fresh: `tests/glb-canonicalize.test.js` + `tests/rig-report.test.js` + `tests/api/forge-anigen.test.js` + `tests/api/forge-anigen-request.test.js` + `tests/api/forge-tiers.test.js` together (722/722); `pytest -q` in `workers/rig/` (23 passed), `workers/rig-nonhuman/` (7 passed) and `workers/model-anigen/` (4 passed); `npm run check:vendored` (OK, 37 copies across 42 workers); `npm run audit:docs` (15 findings, same pre-existing set, none naming rig-nonhuman, anigen or this order); `npm run check:rules` exit 0 on every touched file (0 changed, nothing to flag). The STRUCTURE.md fleet row and the three `data/changelog.json` entries from commits `af041fb9d`/`aa3b812da` are all still present. `gcloud run services describe model-rig-nonhuman --region us-central1` and `model-anigen` both still answer "Cannot find service", and `node scripts/read-service-env.mjs` confirms `three-ws-api` still has neither `RIG_NONHUMAN_URL` nor `MODEL_ANIGEN_URL`, so rows 40 and 41 of `_context/production-100-OWNER-ACTIONS.md` and the order's existing "Evolve status" paragraph (both already dated 2026-10-11) needed no edit. The remaining step stays the owner's two GPU deploys; no code changed this pass beyond this log entry.
+- 2026-10-11: order 070 (editable parts: PartCrafter, HoloPart behind the segment lane) built and committed, NOT closed. Task 1: `workers/model-partcrafter/` (image to part-separated GLB, named nodes per part) and `workers/model-holopart/` (completes occluded geometry of a mesh already segmented by SAM/PartCrafter) both built, each MIT, both VAST-AI-Research, LICENSE files read 2026-10-11. Registered `partcrafter` (self-host image lane, `editable_parts: true`, `minParts`/`maxParts` 2-16 via the new `num_parts` option in `api/_lib/forge-options.js`) and `holopart` (mesh-completion mode behind a new `api/forge-holopart.js` endpoint mirroring `api/forge-segment.js`'s SSRF/FBX/seed validation) end to end through `api/_lib/forge-tiers.js`, `api/_providers/gcp.js`, `api/forge.js`/`api/gpt-forge.js`; both opt-in only, never a default or failover target. Task 2: Restyle Studio (`src/restyle.js`/`pages/restyle.html`) gained a Parts panel: `detectParts()` finds multi-top-level-node models, `editRoot()` scopes every material operation (presets, AI restyle, fine-tuning) to the selected part, and a part can be swapped from a file or regenerated through `/api/forge-holopart`, reassembling under its original node name via `swapSelectedPart()`. Task 3: node-name stability (order 078's dependency) is satisfied by HoloPart's own upstream behavior plus `swapSelectedPart()`'s explicit preservation. Task 4: `STRUCTURE.md` fleet rows, `docs/forge.md` engine-grid entry and editable-parts section with curl examples, `data/changelog.json` entry, `npm run build:pages`; both workers' own READMEs were written alongside the worker code.
+ - Found and fixed a real pre-existing defect while verifying the meshopt-decode hard rule against the new HoloPart worker: `workers/model-holopart/Dockerfile` vendors `gltf_meshopt.py` (which shells out to a `gltfpack` binary to decode `EXT_meshopt_compression`) but never staged that binary, so any real meshopt-compressed input would have failed at runtime with "the gltfpack decoder is not installed". The same grep caught an identical gap in `workers/rig-nonhuman/Dockerfile`, built in order 069. Both now stage the pinned, checksum-verified binary via the proven `workers/segment/Dockerfile` multi-stage pattern. `scripts/check-vendored-workers.mjs` gained a mechanical check (every worker vendoring `gltf_meshopt.py` must contain `COPY --from=gltfpack`) confirmed to fail before the fix and pass after, so this class of gap cannot regress silently again.
+ - Passing: `tests/api/forge-partcrafter.test.js` + `tests/api/forge-holopart.test.js` + `tests/forge-options.test.js` together (46/46, including the `editable_parts` catalog-surfacing regression this caught and a `summarizeForgeOptions`/`num_parts` test-coverage regression fixed along the way); `python3 -m py_compile` clean on every new `.py` file in both workers (no local GPU/torch, matching the documented "local tests" step in each worker's README; the full contract test runs at Docker build time); `node scripts/check-vendored-workers.mjs` OK (41 copies across 44 workers, every gltf_meshopt.py vendor now stages gltfpack); `npm run check:rules` exit 0 on every touched file; `npm run audit:docs` 15 findings, same pre-existing set, none naming partcrafter, holopart or this order; a full `npx vitest run` shows 39 failed files / 73 failed tests out of 2372 files / 35322 tests, matching the established pre-existing concurrent-agent drift pattern from orders 067-069 (none of the failing files touch forge, restyle, partcrafter or holopart). DoD line 1 (3+ named parts, scoped retexture) verified in a real browser via Playwright against a synthetic 3-part GLB fixture built with `GLTFExporter.parse([...], ...)` to match the genuinely-top-level-node shape `workers/segment/segment_core.py`'s `build_scene()` produces in production; zero console errors, correctly scoped history label.
+ - Still open, owner (rows 43 and 44 of `_context/production-100-OWNER-ACTIONS.md`): (1) `bash workers/model-partcrafter/stage-assets.sh`, `gcloud builds submit --config workers/model-partcrafter/cloudbuild.yaml --substitutions=SHORT_SHA=manual$(date +%s) .`, `gcloud run services update three-ws-api --region us-central1 --update-env-vars MODEL_PARTCRAFTER_URL=<service-url>`; (2) the same three-step sequence for `workers/model-holopart/` with `MODEL_HOLOPART_URL`; (3) run one real image through PartCrafter and one real part-segmented mesh through HoloPart, confirm DoD line 1 against live output (not just the synthetic fixture), then delete the order file.
+ - Commits: `d3edf5a25` (both workers + the gltfpack Dockerfile fix + the vendored-workers check), `9b1f6ae9b` (Forge API/provider wiring), `eecca6527` (Restyle Studio Parts panel), `a287682e1` (tests), `ac0c35112` (docs, STRUCTURE.md, changelog).
diff --git a/prompts/finish/_context/production-100-OWNER-ACTIONS.md b/prompts/finish/_context/production-100-OWNER-ACTIONS.md
index b79cbf78f..45194cf92 100644
--- a/prompts/finish/_context/production-100-OWNER-ACTIONS.md
+++ b/prompts/finish/_context/production-100-OWNER-ACTIONS.md
@@ -9,7 +9,7 @@ it and, if that unblocks a work order, say which one in the owning pack's `produ
Rows carry the date they were added so staleness is visible. Re-verify a row before acting
on it; some resolve themselves as other work lands.
-## Open rows (2026-10-11 (triage-lane sweep): row 42 added for the MCP client-sessions nullable-column migration, an active production bug; row 1 re-measured, 28 commits behind `main` (up from 5); 2026-10-11 (best3d order 069 close-out): rows 40 and 41 added for the `model-rig-nonhuman` and `model-anigen` first deploys; 2026-10-11 (best3d order 068 close-out): row 39 added for the `model-pixal3d` first deploy; 2026-10-11 (best3d order 067 retry): row 38 added for the `model-trellis2` first deploy; 2026-10-11 (order 404 close-out): row 34 deleted, its premise cleared by merge commit `d77b89178` (the `evolve`/`main` reconciliation) which also carried the `x-mention-reply.js` `chainFor()` fix row 34 called out, verified live with `RUN_LLM_TESTS=1 npx vitest run tests/x-mention-reply.llm.test.js` (16/16 `source: 'model'`, no `buildBody` TypeError); 2026-10-11 (scout lane): row 36 added for the Cloudflare R2 flapping degraded/down pattern; 2026-10-11 (even later): row 24 re-measured again after order 400's ring-leak-scan tiering/lock fix shipped on `evolve`; 2026-10-11 (later): row 35 added for the agent mention-policy migration; 2026-10-11 04:16 UTC: rows 1 and 24 re-measured by the deep triage sweep; row 34 added for the evolve/main branch divergence; 2026-10-10: row 33 added for the xAI MCP e2e key; rows 1, 19 and 24 re-measured by the triage sweep; 2026-10-09: rows 1, 2 and 24 re-measured by the triage sweep, row 31 added for the GPU worker redeploy and row 32 for database table retention; row 30 added for the Grok Bot CLI setup rerun; 2026-10-08: row 26 added for the coin-gated MCP catalog regeneration; rows 1 and 2 re-measured by the triage sweep, row 24 added for the paid Solana RPC credits, row 25 added for the MCP connector OAuth deploy; 2026-09-30: row 23 added for the $THREE receiver wallets; 2026-09-29: row 19 re-measured after the agent-sniper deploy; 2026-09-09: row 19 added, and row 1 corrected because its outage premise no longer holds; rows 1, 2, 3, 9, 13, 15 and 16 re-measured 2026-09-04; the former row 18 was cleared that day and its number is not reused; the 2026-09-02 and 2026-08-09 passes are folded in below)
+## Open rows (2026-10-11 (best3d order 070 close-out): rows 43 and 44 added for the `model-partcrafter` and `model-holopart` first deploys; 2026-10-11 (triage-lane sweep): row 42 added for the MCP client-sessions nullable-column migration, an active production bug; row 1 re-measured, 28 commits behind `main` (up from 5); 2026-10-11 (best3d order 069 close-out): rows 40 and 41 added for the `model-rig-nonhuman` and `model-anigen` first deploys; 2026-10-11 (best3d order 068 close-out): row 39 added for the `model-pixal3d` first deploy; 2026-10-11 (best3d order 067 retry): row 38 added for the `model-trellis2` first deploy; 2026-10-11 (order 404 close-out): row 34 deleted, its premise cleared by merge commit `d77b89178` (the `evolve`/`main` reconciliation) which also carried the `x-mention-reply.js` `chainFor()` fix row 34 called out, verified live with `RUN_LLM_TESTS=1 npx vitest run tests/x-mention-reply.llm.test.js` (16/16 `source: 'model'`, no `buildBody` TypeError); 2026-10-11 (scout lane): row 36 added for the Cloudflare R2 flapping degraded/down pattern; 2026-10-11 (even later): row 24 re-measured again after order 400's ring-leak-scan tiering/lock fix shipped on `evolve`; 2026-10-11 (later): row 35 added for the agent mention-policy migration; 2026-10-11 04:16 UTC: rows 1 and 24 re-measured by the deep triage sweep; row 34 added for the evolve/main branch divergence; 2026-10-10: row 33 added for the xAI MCP e2e key; rows 1, 19 and 24 re-measured by the triage sweep; 2026-10-09: rows 1, 2 and 24 re-measured by the triage sweep, row 31 added for the GPU worker redeploy and row 32 for database table retention; row 30 added for the Grok Bot CLI setup rerun; 2026-10-08: row 26 added for the coin-gated MCP catalog regeneration; rows 1 and 2 re-measured by the triage sweep, row 24 added for the paid Solana RPC credits, row 25 added for the MCP connector OAuth deploy; 2026-09-30: row 23 added for the $THREE receiver wallets; 2026-09-29: row 19 re-measured after the agent-sniper deploy; 2026-09-09: row 19 added, and row 1 corrected because its outage premise no longer holds; rows 1, 2, 3, 9, 13, 15 and 16 re-measured 2026-09-04; the former row 18 was cleared that day and its number is not reused; the 2026-09-02 and 2026-08-09 passes are folded in below)
| # | Action | Size | Unblocks | Detail lives in |
|---|---|---|---|---|
@@ -53,6 +53,8 @@ on it; some resolve themselves as other work lands.
| 40 | **Deploy `model-rig-nonhuman` (the UniRig+SkinTokens non-humanoid rigging worker), then point `RIG_NONHUMAN_URL` at it from `model-rig`** (added 2026-10-11 by the evolve loop, best3d order 069, built and tested same day). `workers/rig-nonhuman/` is complete (Dockerfile, `cloudbuild.yaml` pinned to `three-ws-build@`, its own test suite passing locally with no GPU). This is an additional GPU service alongside the existing `model-rig`, not a replacement, so budget it against the per-region L4 quota in `docs/ops/gcp-credits-plan.md` before deploying. This is a production deploy (new Cloud Run GPU service, gate 2), so no agent runs it. Commands, in order (from `workers/rig-nonhuman/README.md`): `bash workers/rig-nonhuman/stage-assets.sh` (one time, stages UniRig's skeleton checkpoint, SkinTokens' two checkpoints, and the Qwen3-0.6B config/tokenizer SkinTokens was fine-tuned from), then `gcloud builds submit --config workers/rig-nonhuman/cloudbuild.yaml --substitutions=SHORT_SHA=manual$(date +%s) .` from the repo root, then `gcloud run services update model-rig --region us-central1 --update-env-vars RIG_NONHUMAN_URL=$(gcloud run services describe model-rig-nonhuman --region us-central1 --format='value(status.url)')` (never `--set-env-vars`, which replaces the whole set). `RIG_NONHUMAN_URL` is unset by default, so until that last command runs, 100% of `model-rig` traffic stays on the pre-existing humanoid-only path; deploying the new service alone changes nothing. After wiring, run one real non-humanoid mesh (a creature or prop) through `POST /api/forge?action=rig` and confirm the response GLB is skinned with a valid skeleton. | one weight stage + one build + one env update + one test generation | Closes best3d order 069 task 1; lets `model-rig` rig creatures, props and vehicles instead of leaving them unrigged | [069-best3d-03-rig-anything.md](../069-best3d-03-rig-anything.md), [best3d-PROGRESS.md](best3d-PROGRESS.md) |
| 41 | **Deploy `model-anigen` (the AniGen rig-on-generation GPU worker), then point `MODEL_ANIGEN_URL` at it** (added 2026-10-11 by the evolve loop, best3d order 069, built and tested same day). `workers/model-anigen/` is complete (Dockerfile, `cloudbuild.yaml` pinned to `three-ws-build@`, its own test suite passing locally with no GPU), and `api/forge.js`/`api/gpt-forge.js`/`api/_lib/forge-tiers.js`/`api/_providers/gcp.js`/`src/forge.js` already register and wire the `anigen` backend end to end (single reference image, opt-in only: never a named default or a failover target, catalog `rigged: true` flag, poll response `rigged`/`joint_count` fields, "Rig on generate" engine-grid label). Confirmed live today: `gcloud run services describe model-anigen --region us-central1` answers "Cannot find service", and `three-ws-api`'s env has no `MODEL_ANIGEN_URL`, so the lane reports `backend_unconfigured` today. This is a production deploy (new Cloud Run GPU service, gate 2), so no agent runs it. Commands, in order (from `workers/model-anigen/README.md`): `bash workers/model-anigen/stage-assets.sh` (one time, stages the AniGen checkpoints under `gs://three-ws-model-weights/anigen/`), then `gcloud builds submit --config workers/model-anigen/cloudbuild.yaml --substitutions=SHORT_SHA=manual$(date +%s) .` from the repo root, then `gcloud run services update three-ws-api --region us-central1 --update-env-vars MODEL_ANIGEN_URL=$(gcloud run services describe model-anigen --region us-central1 --format='value(status.url)')` (never `--set-env-vars`). After that, run one real single-image generation with `backend: "anigen"` and confirm the `forge_creations` row's result reports `rigged: true` and a `joint_count`, and that the returned GLB is a valid, already-skinned asset. | one weight stage + one build + one env update + one test generation | Closes best3d order 069 task 2; makes the "Rig on generate" lane (mesh, skeleton and skin weights in one pass, no vendor cost) live in the Forge engine grid | [069-best3d-03-rig-anything.md](../069-best3d-03-rig-anything.md), [best3d-PROGRESS.md](best3d-PROGRESS.md) |
+| 43 | **Deploy `model-partcrafter` (the PartCrafter image-to-parts GPU worker), then point `MODEL_PARTCRAFTER_URL` at it** (added 2026-10-11 by the evolve loop, best3d order 070, built and tested same day). `workers/model-partcrafter/` is complete (Dockerfile, `cloudbuild.yaml` pinned to `three-ws-build@`, its own test suite passing locally with no GPU), and `api/forge.js`/`api/gpt-forge.js`/`api/_lib/forge-tiers.js`/`api/_lib/forge-options.js`/`api/_providers/gcp.js`/`src/forge.js` already register and wire the `partcrafter` backend end to end (single reference image, `num_parts` 2-16 range option, catalog `editable_parts: true` flag, poll response `parts`/`part_count`/`manifest_url` fields, opt-in only: never a named default or a failover target in either direction). Confirmed live today: `gcloud run services describe model-partcrafter --region us-central1` answers "Cannot find service", and `three-ws-api`'s env has no `MODEL_PARTCRAFTER_URL`, so the lane reports `backend_unconfigured` today. This is a production deploy (new Cloud Run GPU service, gate 2), so no agent runs it. Commands, in order (from `workers/model-partcrafter/README.md`): `bash workers/model-partcrafter/stage-assets.sh` (one time, stages the PartCrafter checkpoints), then `gcloud builds submit --config workers/model-partcrafter/cloudbuild.yaml --substitutions=SHORT_SHA=manual$(date +%s) .` from the repo root, then `gcloud run services update three-ws-api --region us-central1 --update-env-vars MODEL_PARTCRAFTER_URL=$(gcloud run services describe model-partcrafter --region us-central1 --format='value(status.url)')` (never `--set-env-vars`). After that, run one real single-image generation with `backend: "partcrafter"` and confirm the result reports ≥2 named parts and the GLB's top-level nodes match them. | one weight stage + one build + one env update + one test generation | Closes best3d order 070 task 1 (image lane half); lets Restyle Studio's part picker work against real self-hosted output instead of only the segment lane | [070-best3d-04-editable-parts.md](../070-best3d-04-editable-parts.md), [best3d-PROGRESS.md](best3d-PROGRESS.md) |
+| 44 | **Deploy `model-holopart` (the HoloPart occluded-part completion GPU worker), then point `MODEL_HOLOPART_URL` at it** (added 2026-10-11 by the evolve loop, best3d order 070, built and tested same day). `workers/model-holopart/` is complete (Dockerfile now stages the `gltfpack` binary gltf_meshopt.py's decode path needs, fixed this same pass; `cloudbuild.yaml` pinned to `three-ws-build@`; its own test suite passing locally with no GPU), and `api/forge-holopart.js`/`api/_providers/gcp.js`/`src/restyle.js` already register and wire the `holopart` mode end to end (SSRF-checked `mesh_url`, FBX rejected, seed validated, Restyle Studio's "Regenerate part" action polls it and reassembles the result under the original node name). Confirmed live today: `gcloud run services describe model-holopart --region us-central1` answers "Cannot find service", and `three-ws-api`'s env has no `MODEL_HOLOPART_URL`, so `/api/forge-holopart` answers 503 `unconfigured` today. This is a production deploy (new Cloud Run GPU service, gate 2), so no agent runs it. Commands, in order (from `workers/model-holopart/README.md`): `bash workers/model-holopart/stage-assets.sh` (one time, stages the HoloPart checkpoints), then `gcloud builds submit --config workers/model-holopart/cloudbuild.yaml --substitutions=SHORT_SHA=manual$(date +%s) .` from the repo root, then `gcloud run services update three-ws-api --region us-central1 --update-env-vars MODEL_HOLOPART_URL=$(gcloud run services describe model-holopart --region us-central1 --format='value(status.url)')` (never `--set-env-vars`). After that, POST a real part-segmented `mesh_url` (e.g. the segment worker's own output) to `/api/forge-holopart` and confirm the polled result's `part_count` matches the input and each part's occluded geometry is filled in. | one weight stage + one build + one env update + one test generation | Closes best3d order 070 task 1 (completion lane half); lets Restyle Studio's "Regenerate part" button produce a real filled-in part instead of only supporting manual swap | [070-best3d-04-editable-parts.md](../070-best3d-04-editable-parts.md), [best3d-PROGRESS.md](best3d-PROGRESS.md) |
| 42 | **Apply the MCP client-sessions nullable-column migration** (added 2026-10-11 by the evolve triage-lane sweep). `api/_lib/migrations/20261015010000_mcp_client_sessions_client_nullable.sql` drops the `NOT NULL` constraint on `mcp_client_sessions.client` and `mcp_client_daily.client`. This is an ACTIVE production bug, not routine deploy lag: `git merge-base --is-ancestor <mcp-clients.js commit> 66a5e93bd` confirms `api/_lib/mcp-clients.js` (the code that inserts session/day rows and can leave `client` unset for some call shapes) is already deployed and running today, so every write that violates the constraint is silently failing right now (lost MCP session/day analytics rows, not a user-facing error since the insert is best-effort). `npm run db:status` lists it as the one pending migration. Command: `node scripts/apply-migrations.mjs --apply --file 20261015010000_mcp_client_sessions_client_nullable.sql` (or `npm run db:migrate`, which applies every pending migration immediately with no dry run). | one migration apply | Stops silent MCP analytics data loss | [docs/ops/production-log-triage.md](../../docs/ops/production-log-triage.md) |
Deleted 2026-09-01: the former row 5 (`WALLET_ENCRYPTION_KEY_PREVIOUS` policy), which