test(translator): fix the relative imports of the relocated deferred-finish test

#11940 moved tests/unit/translator/openai-to-claude-trailing-usage.test.ts one level
up so a collector would run it, but kept the ../../../ import path from the old
directory, so the file failed to load and painted Unit Tests fast-path (3/4) red on
every PR since a94fe23e89. The path now matches its new location (5/5 pass).

.env.example

@@ -3023,3 +3023,11 @@ QUOTA_STORE_DRIVER=sqlite
 # corpus-aware retrieval. Higher values keep more index entries hot.
 # Used by: src/lib/localCorpus/configured.ts
 # OMNIROUTE_CORPUS_CACHE_SIZE=5
+
+# Service-worker cache-busting id for the PWA shell (#11779). NEXT_PUBLIC_SW_BUILD_ID is
+# derived at build time from OMNIROUTE_SW_BUILD_ID, then SOURCE_VERSION (set by some PaaS
+# builders), then the git SHA — override only when the build cannot see git. Used by:
+# next.config.mjs, scripts/build/assembleStandalone.mjs, src/shared/components/PwaRegister.tsx.
+#OMNIROUTE_SW_BUILD_ID=2026-08-28T12-00-00
+#SOURCE_VERSION=abcdef0123456789
+#NEXT_PUBLIC_SW_BUILD_ID=abcdef0123456789

.github/workflows/ci.yml

@@ -609,13 +609,24 @@ jobs:
     # Dynamic runner: when the release captain flips the USE_VPS_RUNNER repo var to
     # 'true' (scripts/vps/release-runner-up.sh does it after the self-hosted VM is
     # online), the heavy jobs run on the dedicated 32-core VPS runners (label
-    # omni-release) instead of queueing on the 20-concurrent-job hosted pool.
+    # omni-build) instead of queueing on the 20-concurrent-job hosted pool.
     # Safety: fork PRs NEVER reach the self-hosted runner — the expression falls
     # back to ubuntu-latest unless the PR head repo is this repository (push /
     # dispatch events are own-origin by definition). Any failure path (VM down,
     # var unset/false) also falls back to ubuntu-latest.
-    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository)) && fromJSON('["self-hosted","omni-release"]') || 'ubuntu-latest' }}
+    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository)) && fromJSON('["self-hosted","omni-build"]') || 'ubuntu-latest' }}
     needs: changes
+    # The .113 pool runs ONE next-build with room to spare and two at the edge: the
+    # box has 31 GB and a single next-build peaks at 14–16 GB RSS. On 2026-08-28
+    # 13:50Z the kernel OOM-killed main's build while a PR build ran beside it
+    # (five Build jobs had been queued by a burst of PRs). Two lanes: main keeps
+    # its own so a release is never queued behind PR traffic; PR builds serialize
+    # among themselves. GitHub keeps one running + one pending per group and
+    # CANCELS older pendings — a cancelled PR build is re-runnable; a dead main
+    # build costs the publish its artefact and a 40-minute rebuild that OOMs.
+    concurrency:
+      group: heavy-build-${{ github.ref == 'refs/heads/main' && 'main' || 'pr' }}
+      cancel-in-progress: false
     if: ${{ github.event_name != 'pull_request' || (needs.changes.outputs.code == 'true' && github.event.pull_request.draft == false) }}
     steps:
       - uses: actions/checkout@v7
@@ -649,14 +660,14 @@ jobs:
         # Keep standalone/node_modules intact: package/electron jobs consume the
         # Next-traced standalone tree and must not replace it with root node_modules.
         run: |
-          tar -czf /tmp/e2e-build.tar.gz \
+          tar -czf "$RUNNER_TEMP/e2e-build.tar.gz" \
             --exclude='.build/next/cache' \
             .build/next
       - name: Upload Next.js build for downstream jobs
         uses: actions/upload-artifact@v7
         with:
           name: next-build
-          path: /tmp/e2e-build.tar.gz
+          path: ${{ runner.temp }}/e2e-build.tar.gz
           retention-days: 1
 
   package-artifact:
@@ -679,10 +690,14 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: next-build
-          path: /tmp/
+          # Workspace-relative on purpose: the matrix below includes windows-latest, whose
+          # default shell is pwsh, where $RUNNER_TEMP is empty (it is $env:RUNNER_TEMP) —
+          # #11896's first cut broke the Electron smoke on exactly that. A relative path
+          # works in bash and pwsh alike; hosted workspaces are ephemeral.
+          path: next-build-artifact
       - name: Extract Next.js build artifact
         run: |
-          tar -xzf /tmp/e2e-build.tar.gz
+          tar -xzf next-build-artifact/e2e-build.tar.gz
       # build:cli consumes the downloaded .build/next standalone artifact and assembles dist/;
       # it only rebuilds if the downloaded standalone artifact is missing.
       - run: npm run build:cli
@@ -770,10 +785,14 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: next-build
-          path: /tmp/
+          # Workspace-relative on purpose: the matrix below includes windows-latest, whose
+          # default shell is pwsh, where $RUNNER_TEMP is empty (it is $env:RUNNER_TEMP) —
+          # #11896's first cut broke the Electron smoke on exactly that. A relative path
+          # works in bash and pwsh alike; hosted workspaces are ephemeral.
+          path: next-build-artifact
       - name: Extract Next.js build artifact
         run: |
-          tar -xzf /tmp/e2e-build.tar.gz
+          tar -xzf next-build-artifact/e2e-build.tar.gz
       - name: Install Electron dependencies
         working-directory: electron
         run: npm install --no-audit --no-fund
@@ -1233,10 +1252,14 @@ jobs:
         uses: actions/download-artifact@v8
         with:
           name: next-build
-          path: /tmp/
+          # Workspace-relative on purpose: the matrix below includes windows-latest, whose
+          # default shell is pwsh, where $RUNNER_TEMP is empty (it is $env:RUNNER_TEMP) —
+          # #11896's first cut broke the Electron smoke on exactly that. A relative path
+          # works in bash and pwsh alike; hosted workspaces are ephemeral.
+          path: next-build-artifact
       - name: Extract Next.js build artifact
         run: |
-          tar -xzf /tmp/e2e-build.tar.gz
+          tar -xzf next-build-artifact/e2e-build.tar.gz
       # WS4.1: duration-balanced shards (LPT over config/quality/e2e-timings.json).
       # Measured skew of plain --shard was 14× (24m47s vs 1m47s) — E2E was the CI
       # critical path. The balancer self-verifies completeness and exits non-zero on

.github/workflows/nightly-release-green.yml

@@ -68,7 +68,7 @@ jobs:
     # this runs on the dedicated VPS runner — clean env (no operator OMNIROUTE_API_KEY,
     # no local noauth CLIs => zero machine-specific false positives) and no contention.
     # Nightly cron normally finds the var false (VM off) and falls back to hosted.
-    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && fromJSON('["self-hosted","omni-release"]')) || 'ubuntu-latest' }}
+    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && fromJSON('["self-hosted","omni-build"]')) || 'ubuntu-latest' }}
     env:
       JWT_SECRET: ci-nightly-secret-with-sufficient-length-for-validation
       API_KEY_SECRET: ci-nightly-api-key-secret-long
@@ -217,7 +217,7 @@ jobs:
     # On a push, only run for a push to main — a push to release/* is handled by
     # release-green above. Schedule/dispatch always run (they also sweep main).
     if: ${{ github.event_name != 'push' || github.ref_name == 'main' }}
-    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && fromJSON('["self-hosted","omni-release"]')) || 'ubuntu-latest' }}
+    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && fromJSON('["self-hosted","omni-build"]')) || 'ubuntu-latest' }}
     env:
       JWT_SECRET: ci-nightly-secret-with-sufficient-length-for-validation
       API_KEY_SECRET: ci-nightly-api-key-secret-long

.github/workflows/npm-publish.yml

@@ -23,11 +23,12 @@ on:
           - next
           - historic
       publish_mode:
-        description: "staged = npm stage publish (owner approves with 2FA after the staged boot-verify); direct = legacy immediate publish (emergency fallback only)"
+        description: "auto = publish through npm Trusted Publishing (OIDC, no token, no 2FA prompt — the default); staged = npm stage publish (owner approves with 2FA); direct = legacy token publish (emergency fallback only)"
         required: false
-        default: "staged"
+        default: "auto"
         type: choice
         options:
+          - auto
           - staged
           - direct
   workflow_call:
@@ -62,7 +63,7 @@ jobs:
     # mid-"Creating an optimized production build" while v3.8.48 had still fit in 16min.
     # This job never runs on `pull_request`, so the fork-safety clause is always true here;
     # it is kept verbatim so the expression stays greppable against ci.yml.
-    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository)) && fromJSON('["self-hosted","omni-release"]') || 'ubuntu-latest' }}
+    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository)) && fromJSON('["self-hosted","omni-build"]') || 'ubuntu-latest' }}
     outputs:
       version: ${{ steps.resolve.outputs.version }}
       tag: ${{ steps.resolve.outputs.tag }}
@@ -204,8 +205,11 @@ jobs:
             exit 0
           fi
           RUN=""
+          # $RUNNER_TEMP, never /tmp: on the .113 pool /tmp is a 12 GB tmpfs (RAM). Parking
+          # this 1.3 GB artefact there took 27–32 min of the 76-min publish job — the
+          # same bytes upload from disk in 2 min. RUNNER_TEMP is per-runner and on disk.
           for candidate in $CANDIDATES; do
-            if gh run download "$candidate" --repo "$REPO" --name next-build --dir /tmp/next-build 2>/dev/null; then
+            if gh run download "$candidate" --repo "$REPO" --name next-build --dir "$RUNNER_TEMP/next-build" 2>/dev/null; then
               RUN="$candidate"
               break
             fi
@@ -215,8 +219,8 @@ jobs:
             echo "::notice::none of the candidate runs still carries next-build (1-day retention) — falling back to a full build"
             exit 0
           fi
-          tar -xzf /tmp/next-build/e2e-build.tar.gz -C .
-          rm -rf /tmp/next-build
+          tar -xzf "$RUNNER_TEMP/next-build/e2e-build.tar.gz" -C .
+          rm -rf "$RUNNER_TEMP/next-build"
           if [ -f .build/next/standalone/server.js ]; then
             echo "✅ standalone tree restored from CI run $RUN — build:cli will skip next build"
           else
@@ -404,8 +408,34 @@ jobs:
           fi
           npm --version
 
+      # Trusted Publishing (OIDC): npm mints a short-lived credential for THIS run from
+      # GitHub's id-token — no NPM_TOKEN secret, no 2FA prompt, provenance included, and
+      # it is the bypass npm sanctions now that tokens which skip 2FA are being retired
+      # (gh.io/npm-gat-bypass2fa-deprecation). Requires the package's Trusted Publisher to
+      # be configured on npmjs.com (owner: diegosouzapw/OmniRoute, workflow
+      # npm-publish.yml) and a github-hosted runner — which is why this job exists.
+      # Without that configuration `npm publish` fails with ENEEDAUTH: re-dispatch with
+      # publish_mode=staged or direct. Automatic publishing was the flow up to v3.8.48;
+      # v3.8.49 moved to staged (WS1.3) to keep a leaked token from publishing alone —
+      # OIDC gives the same guarantee without the manual approve.
+      - name: Publish to npm (Trusted Publishing / OIDC — automatic)
+        if: github.event_name != 'workflow_dispatch' || inputs.publish_mode == 'auto'
+        env:
+          VERSION: ${{ needs.publish.outputs.version }}
+          TAG: ${{ needs.publish.outputs.tag }}
+        run: |
+          set -euo pipefail
+          TARBALL="omniroute-${VERSION}.tgz"
+          test -f "$TARBALL" || { echo "tarball $TARBALL did not arrive from the publish job" >&2; ls -la; exit 1; }
+          # Deliberately NO NODE_AUTH_TOKEN in this step: npm >= 11.5 detects the GitHub
+          # OIDC token itself. Always pass --tag explicitly (defense in depth: an older
+          # VERSION can never claim `@latest`).
+          npm publish "$TARBALL" --provenance --access public --tag "$TAG" --ignore-scripts
+          echo "✅ Published omniroute@$VERSION (dist-tag=$TAG) via Trusted Publishing"
+
       - name: Publish to npm (staged — owner approves with 2FA)
-        if: github.event_name != 'workflow_dispatch' || inputs.publish_mode != 'direct'
+        # Only on an explicit request now: Trusted Publishing below is the default.
+        if: github.event_name == 'workflow_dispatch' && inputs.publish_mode == 'staged'
         env:
           VERSION: ${{ needs.publish.outputs.version }}
           TAG: ${{ needs.publish.outputs.tag }}

changelog.d/features/npm-trusted-publishing-oidc.md

@@ -0,0 +1,4 @@
+- The npm publish is automatic again, through npm Trusted Publishing (OIDC): the hosted
+  `stage-npm` job publishes with a short-lived credential minted from GitHub's id-token —
+  no `NPM_TOKEN`, no 2FA prompt, provenance attached. `publish_mode=staged` (owner
+  approves with 2FA) and `direct` (token) remain available on `workflow_dispatch`.

changelog.d/fixes/11866-alibaba-allowlist-test-timebomb.md

@@ -0,0 +1,5 @@
+- Fixed the Alibaba free-tier allowlist test that went red on its own once the
+  shipped catalog's `validUntil` (2026-08-27) passed, leaving every PR and `main`
+  with a failing `Unit Tests (1/8)`. The test now builds its own packs with dates
+  it controls, and covers the expired-pack fallback that production has actually
+  been serving.

changelog.d/fixes/11879-xai-xhigh-reasoning-effort.md

@@ -0,0 +1 @@
+- **fix(providers):** xAI `reasoning_effort: "xhigh"` now reaches grok-4.6+ instead of being silently clamped to `"high"` ([#11879](https://github.com/diegosouzapw/OmniRoute/pull/11879)) — thanks @NoxzRCW

changelog.d/fixes/11880-provider-icon-prototype-collision.md

@@ -0,0 +1 @@
+- **fix(dashboard):** the providers page no longer crashes into the error boundary when a provider id collides with an `Object.prototype` member (`constructor`, `__proto__`); icon lookups are own-property guarded ([#11880](https://github.com/diegosouzapw/OmniRoute/pull/11880)) — thanks @NoxzRCW

changelog.d/fixes/11881-skills-shorthand-tool-schema.md

@@ -0,0 +1 @@
+- **fix(skills):** injected skill tools declared in shorthand (`{"content": "string"}`) now forward valid JSON Schema, unblocking providers that validate tool schemas strictly such as Zhipu GLM on the Console Go tier ([#11881](https://github.com/diegosouzapw/OmniRoute/pull/11881)) — thanks @NoxzRCW

changelog.d/fixes/11882-simulate-route-step-warnings.md

@@ -0,0 +1 @@
+- **fix(api):** `POST /api/playground/simulate-route` now surfaces `combo-ref` and `provider-wildcard` persisted combo steps with a specific warning (naming the referenced combo, or the unresolved `provider/model` wildcard) instead of folding them into a generic "unsupported step" count; a `provider-wildcard` step is also now included as an unresolved target so the operator can see it is in the route (ported from [#11882](https://github.com/diegosouzapw/OmniRoute/pull/11882) — thanks @NoxzRCW).

changelog.d/fixes/11883-openai-to-claude-trailing-usage.md

@@ -0,0 +1 @@
+- **fix(translator):** the streaming OpenAI→Claude translator keeps upstream usage, including prompt-cache tokens, when it arrives on a trailing `choices: []` chunk (Fireworks and any upstream using `stream_options.include_usage`) ([#11883](https://github.com/diegosouzapw/OmniRoute/pull/11883)) — thanks @NoxzRCW

changelog.d/fixes/11915-openai-to-claude-usage-emission-ordering.md

@@ -0,0 +1 @@
+- **fix(sse):** the OpenAI→Claude stream translator now defers the terminal `message_delta`/`message_stop` emission until the real usage block has arrived (or a genuine end-of-stream flush forces it) instead of emitting it immediately on `finish_reason` — previously, when the trailing usage-only chunk (`{"choices":[],"usage":{...}}`) arrived *after* the `finish_reason` chunk (the normal order for Fireworks/vLLM/Together and other `stream_options.include_usage` upstreams), the client-visible `message_delta` still carried stale/zero usage even though `state.usage` was internally corrected too late to matter (ported from [#11915](https://github.com/diegosouzapw/OmniRoute/pull/11915) — thanks @HouMinXi).

changelog.d/fixes/11916-strip-store-openai-compatible-responses.md

@@ -0,0 +1 @@
+- **fix(sse):** `stripStore()` now forces `store=false` for stateless OpenAI-compatible Responses-API targets (unless the connection has explicitly opted in via `providerSpecificData.openaiStoreEnabled`), instead of only handling the `openai`/`agentrouter` cases — a client-supplied `store` value previously passed through untouched to backends that don't actually persist responses server-side ([#11916](https://github.com/diegosouzapw/OmniRoute/pull/11916) — thanks @HouMinXi).

changelog.d/fixes/11918-custom-node-canonical-prefix.md

@@ -0,0 +1 @@
+- **fix(models):** custom provider-node models (synced, custom, and alias-backed) now appear under their configured prefix in the unified catalog when the operator's model-id prefix mode is canonical, instead of being dropped whenever alias-inclusion was otherwise disabled ([#11918](https://github.com/diegosouzapw/OmniRoute/pull/11918) — thanks @HouMinXi).

changelog.d/fixes/11919-static-catalog-authoritative-suppression.md

@@ -0,0 +1 @@
+- **fix(models):** the unified model catalog now suppresses stale static registry models (including effort-tier variants) for any provider whose active connection has an authoritative live synced catalog, not only providers already using exclusive-synced-listing — a connection with `providerUsesAuthoritativeLiveCatalog` previously kept serving both the live-synced models and the stale static rows side by side ([#11919](https://github.com/diegosouzapw/OmniRoute/pull/11919) — thanks @HouMinXi).

changelog.d/fixes/11920-provider-icon-prototype-guards.md

@@ -0,0 +1 @@
+- **fix(ui):** `ProviderIcon`'s three remaining unguarded lookups (`PROVIDER_ICON_ALIASES`, `LOCAL_SVG_ALIASES`, `THEMED_SVGS`) now use `Object.hasOwn()` own-property checks — a provider id such as `constructor` or `__proto__` previously resolved through the prototype chain instead of falling through to the unknown-provider CDN fallback (`getLobeProviderIcon()` itself was already guarded by [#11880](https://github.com/diegosouzapw/OmniRoute/pull/11880)); `ProviderPageHeader`'s `color` field is also now optional, matching the rest of the component's defensive typing (ported from [#11920](https://github.com/diegosouzapw/OmniRoute/pull/11920) — thanks @HouMinXi).

changelog.d/fixes/v3851-sweep-reds.md

@@ -0,0 +1,7 @@
+- Drained the reds every PR against `release/v3.8.51` was born with: documented the
+  three service-worker build-id variables, registered the six retirement/tunnel tests
+  with the mutation gate, approved `eslint-plugin-react-hooks` in the dependency
+  allowlist, added the six `combo.sort.*` strings to `vi` and `pt-BR`, pointed the
+  ChatGPT Web doc at the real migration-168 test, worded the g4f hint around the
+  member key, and realigned four tests to the retired-provider catalog and the
+  legacy-schema fixtures the retirement migrations touch.

changelog.d/maintenance/11866-config-expiry-time-bomb-test.md

@@ -0,0 +1,4 @@
+- Added a unit test that fails seven days before any dated pack under `config/`
+  (`validUntil` and sibling keys) lapses, naming the file and key. The Alibaba
+  free-tier pack expired on 2026-08-27 and turned every PR red the next morning
+  with no commit involved; renewal now happens on someone's terms, not the clock's.

changelog.d/maintenance/11878-check-workflows-provenance-self-hosted.md

@@ -0,0 +1,3 @@
+- `check:workflows` now fails (under `--strict`/`--ratchet`) when any job routed to a
+  self-hosted runner publishes with `--provenance` — npm rejects that with `422` at the
+  registry, which in v3.8.50 only surfaced after the tag and Docker images were public.

changelog.d/maintenance/11892-runner-janitor-act-not-advise.md

@@ -0,0 +1,5 @@
+- `scripts/ops/runner-janitor.sh` now proves a path is idle with one `lsof`
+  snapshot and removes stale leftovers itself (tmpfs after 3 h — it is RAM — disk
+  after 24 h), kills orphan `next-build` processes, prunes checkouts of stopped
+  runners, and alerts on memory pressure; `--dry-run` shows exactly what it would
+  do. `docs/ops/RUNNER_BOX.md` reconciled to the measured box (31 GB, 10 listeners).

changelog.d/maintenance/11896-ci-artifact-download-to-disk.md

@@ -0,0 +1,5 @@
+- The `next-build` artefact (1.3 GB) is now written and read under `$RUNNER_TEMP`
+  (per-runner, on disk) instead of `/tmp`, which on the self-hosted pool is a
+  12 GB tmpfs in RAM. Landing it there took 27–32 of the publish job's 76 minutes,
+  and the fixed `/tmp/e2e-build.tar.gz` name let E2E jobs on different runners
+  overwrite each other's download.

changelog.d/maintenance/11897-ci-heavy-build-lane.md

@@ -0,0 +1,3 @@
+- The CI `build` job now runs in two concurrency lanes — `main` and pull requests —
+  so a release build is never queued behind (or OOM-killed beside) PR builds on the
+  self-hosted pool, which holds one `next-build` comfortably and two at the edge.

changelog.d/maintenance/ci-omni-build-runner-label.md

@@ -0,0 +1,4 @@
+- Every CI job that runs a `next build` (`build`, the npm `publish`, both release-green
+  validations) now targets the `omni-build` runner label, which only two of the eight
+  self-hosted runners carry. The box holds one build comfortably and two at the edge; a
+  third now queues on GitHub instead of being OOM-killed by the kernel.

config/quality/dependency-allowlist.json

@@ -2,7 +2,8 @@
   "_comment": "Allowlist anti-slopsquatting (check-deps.mjs). Toda dep nova exige adicao EXPLICITA aqui apos verificar que e legitima.",
   "_justifications": {
     "@testing-library/dom": "Peer dep obrigatoria de @testing-library/react v16 (adicionada no PR #11224); Refs #9985.",
-    "@testing-library/user-event": "Utilitario oficial do ecossistema testing-library para testes de UI (adicionada no PR #11224); Refs #9985."
+    "@testing-library/user-event": "Utilitario oficial do ecossistema testing-library para testes de UI (adicionada no PR #11224); Refs #9985.",
+    "eslint-plugin-react-hooks": "React Hooks lint rules (set-state-in-effect, immutability, refs, purity) pinned at 7.0.1 by the release/v3.8.51 cycle; the 224 findings it raised are tracked in #11924. Refs #11924."
   },
   "allowed": [
     "@atjsh/llmlingua-2",
@@ -48,8 +49,8 @@
     "clsx",
     "commander",
     "concurrently",
-    "cross-env",
     "cron-parser",
+    "cross-env",
     "csv-stringify",
     "ctrf",
     "dompurify",
@@ -60,6 +61,7 @@
     "esbuild",
     "eslint",
     "eslint-config-next",
+    "eslint-plugin-react-hooks",
     "eslint-plugin-sonarjs",
     "express",
     "fast-check",
@@ -102,9 +104,9 @@
     "node-loader",
     "node-machine-id",
     "omniglyph",
+    "onnxruntime-node",
     "open",
     "opencode-ai",
-    "onnxruntime-node",
     "ora",
     "parse5",
     "pino",
@@ -131,10 +133,10 @@
     "tailwind-merge",
     "tailwindcss",
     "tls-client-node",
-    "turndown",
-    "turndown-plugin-gfm",
     "tsup",
     "tsx",
+    "turndown",
+    "turndown-plugin-gfm",
     "type-coverage",
     "typescript",
     "typescript-eslint",

config/quality/file-size-baseline.json

@@ -45,6 +45,7 @@
   "_rebaseline_2026_07_03_v3844_ipfilter_release_green": "testFrozen bumps: models-catalog-route 1507->1600, perplexity-web 959->999, route-edge-coverage 1234->1241 (last is my #5975 comment +7). v3.8.44 cycle drift measured on release tip 32e4c906e during the #6131/#5975 release-green rebaseline. Inherited from the merge burst (Quality Ratchet does not run on PR->release fast-gates). route-edge-coverage +7 is my #5975 test comment; the rest is parallel-session drift. Tighten via --update next cycle.",
   "_rebaseline_2026_07_03_v3844_residual_release_green": "Residual file-size drift on tip 716041223: providerLimits.ts 955->982 + accountFallback.ts 1790->1864 (production god-files grown by parallel-session merges e.g. #6128; ideally DECOMPOSE not rebaseline, tracked as debt) + sse-auth.test.ts 1553->1600. None mine.",
   "_comment": "Catraca de tamanho (check-file-size.mjs). frozen so pode encolher; arquivos novos <= cap. --update ratcheta.",
+  "_rebaseline_2026_08_28_mergebatch_v3851_noxzrcw_skills_pipeline_drift": "/merge-batch 2026-08-28 (v3.8.51): tests/integration/skills-pipeline.test.ts already measured 1008->1009 (gate) on the pure release/v3.8.51 tip before boarding any PR in this batch (#11883/#11881/#11880/#11879 — none touch this file); pre-existing drift inherited from an earlier already-merged PR, rebaselined here so the gate stays green for this batch.",
   "_rebaseline_2026_07_09_pr6647_winget_claude_detect": "PR #6647 (enjoyer-hub, /implement-prs sync): cliRuntime.ts 1100->1110 (split('\\n').length metric; +10, was already exactly at the 1100 frozen cap). Adds the WinGet-installed Claude Code fallback path (%LOCALAPPDATA%\\Microsoft\\WinGet\\Packages\\Anthropic.ClaudeCode_Microsoft.Winget.Source_8wekyb3d8bbwe\\claude.exe) to getKnownToolPaths() alongside the two sibling Claude Code paths, so WinGet installs are auto-detected without CLI_CLAUDE_BIN. The package folder name (62 chars) forces Prettier's 100-char width to break the path.join call across the full 10-line multi-arg form used elsewhere in this same function for long paths; irreducible without changing the shared getKnownToolPaths() structure. Covered by the PR's own regression test (tests/unit/cli-runtime-detection.test.ts, win32-gated).",
   "_rebaseline_2026_07_03_review_prs_release_green": "Release-green unblock (2026-07-03, /review-prs): the quality.yml fast-gates job was base-red for EVERY PR->release from growth inherited via already-merged PRs on the release tip — no offending PR branch left to fix in-place. Prod frozen raised: ApiManagerPageClient.tsx 3017->3058, OAuthModal.tsx 969->989, cliRuntime.ts 1090->1100, webProvidersA.ts 805->809. Test frozen raised: deepseek-web.test.ts 1081->1092. Real sizes (check-file-size.mjs reported). These stay frozen (cannot grow further); structural shrink tracked under decomposition roadmap #3501; the release captain's rebaseline-at-release supersedes this note. Bundled with the #5695 quick-start test regex fix (multi-line <Link> tolerance) in the same release-green PR.",
   "_rebaseline_2026_07_02_5798_release_green": "Release-green unblock #5798 / PR #5896 (2026-07-02): the quality.yml fast-gates job was base-red for EVERY PR->release (whole queue failing), from growth inherited via already-merged PRs — no offending PR branch left to fix. Prod frozen raised: AddApiKeyModal.tsx 869->905, providerPageHelpers.ts 996->1021, RequestLoggerV2.tsx 1316->1553, src/sse/services/auth.ts 2403->2405, antigravity.ts 1806->1813, base.ts 1502->1536 (1533 inherited + 3 lines from this PR's own typecheck:core fix in resolveBaseUrl), advancedTools.ts 1118->1120, accountFallback.ts 1783->1790, openai-to-kiro.ts 842->853, openai-responses.ts 1035->1092, stream.ts 2710->2727; new-above-cap frozen: webProvidersA.ts 805, tokenHealthCheck.ts 830. Test frozen raised: cc-compatible-provider 1179->1217, translator-openai-to-kiro 999->1088, web-cookie-providers-new 827->845; new-above-cap: response-sanitizer.test.ts 906. These files remain frozen (cannot grow further); the release captain's rebaseline-at-release supersedes this note.",
@@ -195,7 +196,7 @@
     "_rebaseline_pr4613_compatible_provider_groups": "Reconcile #4613 already-merged growth: providers-page-utils.test.ts 1004->1052 (+48, buildCompatibleProviderGroups partition unit test). Fast-gate PR->release does not run check:file-size, so this surfaced post-merge.",
     "tests/integration/chat-pipeline.test.ts": 2077,
     "tests/integration/chatcore-compression-integration.test.ts": 1448,
-    "tests/integration/skills-pipeline.test.ts": 1006,
+    "tests/integration/skills-pipeline.test.ts": 1009,
     "tests/unit/account-fallback-service.test.ts": 2032,
     "tests/unit/adobe-firefly.test.ts": 1477,
     "tests/unit/batch_api.test.ts": 1721,

docs/ops/RELEASE_CHECKLIST.md

@@ -1,12 +1,12 @@
 ---
 title: "Release Checklist"
-version: 3.8.40
-lastUpdated: 2026-06-28
+version: 3.8.51
+lastUpdated: 2026-08-28
 ---
 
 # Release Checklist
 
-> **Last updated:** 2026-06-28 — v3.8.40
+> **Last updated:** 2026-08-28 — v3.8.51
 > Streamlined release flow that leverages Claude Code skills for automation.
 >
 > **Keep the queue/branch green between releases:** see [RELEASE_GREEN.md](./RELEASE_GREEN.md)
@@ -37,7 +37,21 @@ npm run test:e2e           # optional but recommended
 /capture-release-evidences-cc
 ```
 
-## npm Staged Publishing (default since v3.8.49 — WS1.3/D2)
+## npm Trusted Publishing (default since v3.8.51) — staged on request, direct as fallback
+
+`npm-publish.yml` publishes through **npm Trusted Publishing (OIDC)** by default: the
+`stage-npm` job (github-hosted) exchanges GitHub's id-token for a short-lived npm
+credential for that run — no long-lived npm token in the repository secrets, no 2FA prompt, provenance attached.
+That is the bypass npm sanctions now that tokens which skip 2FA are being retired;
+it restores the fully automatic flow the project had up to v3.8.48 while keeping the
+WS1.3 guarantee (a leaked token cannot publish alone — there is no token).
+
+**One-time setup (owner):** npmjs.com → package `omniroute` → Settings → *Trusted
+Publisher* → GitHub: owner `diegosouzapw`, repo `OmniRoute`, workflow `npm-publish.yml`
+(environment: none). Until that exists, the automatic step fails with `ENEEDAUTH`:
+re-dispatch with `publish_mode=staged` (below) or `direct`.
+
+### Staged publishing (on request — `publish_mode=staged`)
 
 The npm-publish workflow no longer publishes directly: it boots the packed tarball
 (`check:pack-boot`) and then runs `npm stage publish` — the exact bytes are parked on

docs/ops/RUNNER_BOX.md

@@ -4,32 +4,66 @@ title: Self-Hosted Runner Box Operations
 
 # Self-Hosted Runner Box Operations (.113 pool)
 
-The self-hosted pool (`self-hosted, omni-release` labels) runs on the 16 GB box at
-`192.168.0.113`. Two failure modes recurred on release days and were, until v3.8.49,
-manual discipline; the **janitor script codifies them** (WS3.3 of the quality plan):
+The self-hosted pool (`self-hosted, omni-release` on all eight runners; `omni-build` on two) runs on the **.113** box.
+Measured 2026-08-28 (v3.8.50 postmortem, Parte III):
 
-1. **Orphaned temp/work dirs** filling the disk → disk-full SQLite errors mid-job.
-2. **>4 concurrent runners** → OOM-killed jobs (8-wide killed jobs twice on the
-   v3.8.47 release day; 4-wide is the proven ceiling).
+| resource  | value                                                            | what it means for scheduling                                                                                                                  |
+| --------- | ---------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------- |
+| RAM / CPU | **31 GB / 32 cores** (was 16 GB when this doc was first written) | one `next-build` peaks at **~14 GB** → 2 concurrent heavy builds saturate the box, 3 take it down (2026-08-28 06:42Z: load 56, two jobs lost) |
+| swap      | 15 GB                                                            | it swapped its way through the v3.8.50 publish; pressure shows in `/proc/pressure/memory`                                                     |
+| `/tmp`    | **12 GB tmpfs = RAM**                                            | anything parked there is memory; leftovers are swept after 3 h                                                                                |
+| disk      | 188 GB                                                           | `_work` checkouts of 8 runners reach ~70 GB with no cap                                                                                       |
+| runners   | **10 listeners**: 8 OmniRoute + OmniHeuris + OmniMind            | all share the memory above                                                                                                                    |
 
 ## Install the janitor (one-time, on the box)
 
 ```bash
-sudo mkdir -p /opt/omniroute-ops
-sudo cp scripts/ops/runner-janitor.sh /opt/omniroute-ops/
-sudo chmod +x /opt/omniroute-ops/runner-janitor.sh
-( sudo crontab -l 2>/dev/null; echo '*/30 * * * * /opt/omniroute-ops/runner-janitor.sh >> /var/log/runner-janitor.log 2>&1' ) | sudo crontab -
+scp scripts/ops/runner-janitor.sh root@192.168.0.113:/opt/omniroute-ops/runner-janitor.sh
+ssh root@192.168.0.113 'chmod +x /opt/omniroute-ops/runner-janitor.sh; apt-get install -y lsof'
+# cron (root): every 30 min, log to /var/log/runner-janitor.log
+*/30 * * * * MAX_ACTIVE_RUNNERS=8 /opt/omniroute-ops/runner-janitor.sh >> /var/log/runner-janitor.log 2>&1
 ```
 
-What it does every 30min: sweeps runner temp leftovers older than 24h, alerts at
-≥85% root-disk usage, and alerts when more than the runner ceiling (default 4, tunable
-via the script's own environment) of `Runner.Listener` processes are up. Alerts land in `/var/log/runner-janitor.log`
-with a non-zero exit (grep for `⚠`).
+`lsof` is required: the janitor proves a path is idle with one snapshot of open
+files before removing it, and without the tool it removes nothing and says so
+(exit 1). Try any change with `--dry-run` first — it prints exactly what it would
+do and touches nothing.
+
+What it does every run: sweeps our own leftovers (`runner-*`, `omniroute-*`,
+`next-build*`, `e2e-build.tar.gz`) after **3 h on tmpfs** and 24 h on disk
+`_work/_temp`; kills a `next-build` older than 75 min (no job runs that long — on
+2026-08-27 one ran 70 min after GitHub had declared its job lost); prunes 48 h-old
+checkouts of runners whose unit is **stopped**; alerts on disk ≥ 85 %, memory PSI
+`full/avg60` ≥ 10 %, and more listeners than `MAX_ACTIVE_RUNNERS` (with an
+omniroute/other breakdown). Exit 1 = attention needed; read the log.
+
+## Runner units: KillMode
+
+The runner's default `KillMode=process` leaves `Runner.Worker → npm → next-build`
+alive when a unit is stopped or restarted — an orphan build keeps eating RAM and
+CPU with no job attached. Every OmniRoute unit carries a drop-in
+(`/etc/systemd/system/actions.runner.diegosouzapw-OmniRoute.<name>.service.d/10-killmode.conf`)
+with `KillMode=mixed`: SIGTERM to the listener first, SIGKILL to the whole cgroup at
+`TimeoutStop`. It takes effect on the unit's next restart — restart **one runner at
+a time, only when idle**, with the idle check and the restart in the same command.
 
 ## Operating rules
 
-- **Ceiling: 4 runners** on the 16 GB box. Runners 5–8 stay STOPPED except for
-  explicit off-peak experiments — never during a release window.
-- Stopping a runner mid-job cancels the job (observed live): `systemctl stop`
-  only when its runner is idle (`Runner.Listener` without a `Runner.Worker` child).
+- **Heavy-build ceiling: 2 at a time — enforced by label.** Every job that runs a
+  `next build` (`ci.yml` `build`, `npm-publish.yml` `publish`, both `nightly-release-green`
+  validations) targets `[self-hosted, omni-build]`, and only **two** runners carry that
+  label (`omniroute-113-5`, `omniroute-113-6`, added through the runners API — no
+  re-registration). The other six keep `omni-release` and take nothing heavy; GitHub
+  queues a third build instead of the kernel killing one. Pair with the `heavy-build-*`
+  concurrency lanes in `ci.yml`. To add capacity, label another runner — never raise
+  the count past what 31 GB holds (one next-build ≈ 14–16 GB).
+- **Never clean `/tmp` or `_work` by hand while any runner is busy.** A
+  check-then-delete with a gap between the two is how a live Build job lost its
+  `_work` on 2026-08-27. The janitor does the check and the removal in one step;
+  let it.
+- Stopping a runner mid-job cancels the job (observed live): `systemctl stop` only
+  when its listener has no `Runner.Worker` child — and do it in one command.
+- Workflows must not park artefacts in `/tmp` (it is RAM). Download to
+  `$RUNNER_TEMP` (on disk, per runner) — the 1.3 GB `next-build` artefact took 27–32
+  minutes to land on the tmpfs and 2 minutes to upload from disk.
 - The `.15` VPS is homologation-only — never runs CI runners.

docs/providers/CHATGPT_WEB.md

@@ -130,4 +130,4 @@ Retirement regression guards live in:
 - `tests/unit/chatgpt-web-runtime-block.test.ts`
 - `tests/unit/chatgpt-web-image-handler-retirement.test.ts`
 - `tests/unit/chatgpt-web-source-retirement.test.ts`
-- `tests/unit/migration-163-retire-chatgpt-web.test.ts`
+- `tests/unit/migration-168-retire-chatgpt-web.test.ts`

docs/reference/ENVIRONMENT.md

@@ -59,6 +59,9 @@ These **must** be set before the first run. Without them, the application will e
 | `API_KEY_SECRET`             | **Yes**              | _(none)_          | `src/lib/db/apiKeys.ts`                            | AES encryption key for API key values at rest in SQLite. Generate with `openssl rand -hex 32`.                                                                                                                                                                                                                                                                     |
 | `INITIAL_PASSWORD`           | **Yes**              | `CHANGEME`        | Bootstrap script                                   | Sets the initial admin dashboard password (matches `.env.example` default — kept obviously insecure to force a change). **Change before first use.** After login, change via Dashboard → Settings → Security.                                                                                                                                                      |
 | `OMNIROUTE_WS_BRIDGE_SECRET` | **Yes** (production) | _(unset)_         | `src/app/api/internal/codex-responses-ws/route.ts` | Shared secret for the internal Codex Responses WebSocket bridge. Authenticates bridge requests between the Electron/browser WS relay and OmniRoute. ⚠️ **REQUIRED in production — when unset, all WS bridge requests are rejected.** Generate with `openssl rand -base64 32`.                                                                                      |
+| `OMNIROUTE_SW_BUILD_ID` | No | _(git SHA)_ | `next.config.mjs`, `scripts/build/assembleStandalone.mjs` | Explicit service-worker cache-busting id for the PWA shell (#11779); first in the resolution chain. |
+| `SOURCE_VERSION` | No | _(unset)_ | `next.config.mjs`, `scripts/build/assembleStandalone.mjs` | Second in the chain — set by PaaS builders (e.g. Heroku-style) as the deployed commit. |
+| `NEXT_PUBLIC_SW_BUILD_ID` | No | _(derived)_ | `src/shared/components/PwaRegister.tsx` | Build-time public value the client uses to register `/sw.js?v=…`; derived from the two above, then the git SHA. |
 | `OMNIROUTE_PEER_STAMP_TOKEN` | No (auto)            | _(auto per boot)_ | `src/server/authz/policies/management.ts`          | Per-process secret proving the trusted peer-IP stamp came from OmniRoute's own HTTP server (`scripts/dev/peer-stamp.mjs`). The authz middleware trusts request locality (loopback/LAN gating of LOCAL_ONLY routes) only when the stamp carries this token. Auto-generated each boot — leave unset; only pin it for multi-process setups that must share the stamp. |
 
 ### Generation Commands

open-sse/config/cliFingerprints.ts

@@ -265,7 +265,7 @@ export function orderHeaders(
  * Apply a CLI fingerprint to headers and body.
  * Returns { headers, bodyString } with the correct ordering.
  */
-function stripInternalBodyFields(body: unknown): unknown {
+export function stripInternalBodyFields(body: unknown): unknown {
   if (!body || typeof body !== "object" || Array.isArray(body)) return body;
 
   const record = body as Record<string, unknown>;

open-sse/executors/base.ts

@@ -30,7 +30,7 @@ import {
   addParamToBlocklist,
   isAutoLearnGloballyEnabled,
 } from "@/lib/db/paramFilters";
-import { applyFingerprint, isCliCompatEnabled } from "../config/cliFingerprints.ts";
+import { applyFingerprint, isCliCompatEnabled, stripInternalBodyFields } from "../config/cliFingerprints.ts";
 import { supportsClaudeMaxEffort, supportsXHighEffort } from "../config/providerModels.ts";
 import { getThinkingBudgetConfig, ThinkingMode } from "../services/thinkingBudget.ts";
 import {
@@ -582,6 +582,8 @@ export class BaseExecutor {
         if (cloned[key] === "") delete cloned[key];
       }
 
+      stripInternalBodyFields(cloned);
+
       return cloned;
     }
 
@@ -1393,6 +1395,7 @@ export class BaseExecutor {
           );
         }
 
+        stripInternalBodyFields(transformedBody);
         let bodyString = JSON.stringify(transformedBody);
 
         const shouldFingerprint =

open-sse/handlers/chatCore.ts

@@ -2808,7 +2808,12 @@ export async function handleChatCore({
     log?.debug?.("PARAMS", `Renamed max_completion_tokens to max_tokens for ${model}`);
   }
 
-  stripStore(translatedBody, provider, targetFormat);
+  stripStore(
+    translatedBody,
+    provider,
+    targetFormat,
+    credentials?.providerSpecificData as Record<string, unknown> | null | undefined
+  );
 
   // Chat clients may send stream_options.include_usage, but OpenAI Responses
   // upstreams (including Azure AI Foundry /responses) reject stream_options.

open-sse/handlers/chatCore/agentRouterProtocol.ts

@@ -26,8 +26,21 @@ export function usesClaudeBridge(
 export function stripStore(
   body: Record<string, unknown>,
   provider: string,
-  targetFormat: string
+  targetFormat: string,
+  providerSpecificData?: unknown
 ): void {
+  if (provider.startsWith("openai-compatible-") && targetFormat === FORMATS.OPENAI_RESPONSES) {
+    const psd =
+      providerSpecificData && typeof providerSpecificData === "object"
+        ? (providerSpecificData as Record<string, unknown>)
+        : undefined;
+    if (psd?.openaiStoreEnabled === true) {
+      return;
+    }
+    body.store = false;
+    return;
+  }
+
   const supportsStore =
     provider === "openai" ||
     (provider === "agentrouter" && targetFormat === FORMATS.OPENAI_RESPONSES);

open-sse/services/tokenExtractionConfig.ts

@@ -173,26 +173,6 @@ const RAW_CONFIGS: TokenExtractionConfig[] = [
     }
   ),
 
-  // ── Volcano Engine Ark Console ───────────────────────────
-  config(
-    "volcengine-console",
-    "Volcano Engine Ark Console",
-    "https://console.volcengine.com/ark/region:cn-beijing/subscription/coding-plan",
-    "https://console.volcengine.com",
-    [
-      { type: "cookie", name: "digest", domain: ".volcengine.com" },
-      { type: "cookie", name: "AccountID", domain: ".volcengine.com" },
-      { type: "cookie", name: "csrfToken", domain: ".volcengine.com" },
-      { type: "cookie", name: "userInfo", domain: ".volcengine.com" },
-    ],
-    "Log in to the Volcano Engine Ark console. The console session is used to discover Agent/Coding Plan API keys and live quota usage.",
-    {
-      cookieDomain: ".volcengine.com",
-      successUrlPattern: /console\.volcengine\.com\/ark/i,
-      pollingConfig: { timeout: 300_000, minLoginTime: 3000 },
-    }
-  ),
-
   // ── Kimi Web ──────────────────────────────────────────────
   config(
     "kimi-web",

open-sse/translator/response/openai-to-claude.ts

@@ -194,50 +194,75 @@ function stopTextBlock(state, results) {
   state.textBlockStarted = false;
 }
 
-// Convert OpenAI stream chunk to Claude format
-export function openaiToClaudeResponse(chunk, state) {
-  if (!chunk || !chunk.choices?.[0]) return null;
+// Harvest the upstream usage block from any chunk, including trailing
+// usage-only chunks that carry `choices: []` (#11817).
+function trackUsageFromChunk(chunk, state) {
+  if (!chunk.usage || typeof chunk.usage !== "object") return;
+  const promptTokens =
+    typeof chunk.usage.prompt_tokens === "number" ? chunk.usage.prompt_tokens : 0;
+  const outputTokens =
+    typeof chunk.usage.completion_tokens === "number" ? chunk.usage.completion_tokens : 0;
 
-  const results = [];
-  const choice = chunk.choices[0];
-  const delta = choice.delta;
+  // Extract cache tokens from prompt_tokens_details
+  const cachedTokens = chunk.usage.prompt_tokens_details?.cached_tokens;
+  const cacheCreationTokens = chunk.usage.prompt_tokens_details?.cache_creation_tokens;
+  const cacheReadTokens = typeof cachedTokens === "number" ? cachedTokens : 0;
+  const cacheCreateTokens = typeof cacheCreationTokens === "number" ? cacheCreationTokens : 0;
 
-  // Track usage from OpenAI chunk if available
-  if (chunk.usage && typeof chunk.usage === "object") {
-    const promptTokens =
-      typeof chunk.usage.prompt_tokens === "number" ? chunk.usage.prompt_tokens : 0;
-    const outputTokens =
-      typeof chunk.usage.completion_tokens === "number" ? chunk.usage.completion_tokens : 0;
+  // input_tokens = prompt_tokens - cached_tokens - cache_creation_tokens
+  // Because OpenAI's prompt_tokens includes all prompt-side tokens
+  const inputTokens = promptTokens - cacheReadTokens - cacheCreateTokens;
 
-    // Extract cache tokens from prompt_tokens_details
-    const cachedTokens = chunk.usage.prompt_tokens_details?.cached_tokens;
-    const cacheCreationTokens = chunk.usage.prompt_tokens_details?.cache_creation_tokens;
-    const cacheReadTokens = typeof cachedTokens === "number" ? cachedTokens : 0;
-    const cacheCreateTokens = typeof cacheCreationTokens === "number" ? cacheCreationTokens : 0;
+  state.usage = {
+    input_tokens: inputTokens,
+    output_tokens: outputTokens,
+  };
 
-    // input_tokens = prompt_tokens - cached_tokens - cache_creation_tokens
-    // Because OpenAI's prompt_tokens includes all prompt-side tokens
-    const inputTokens = promptTokens - cacheReadTokens - cacheCreateTokens;
-
-    state.usage = {
-      input_tokens: inputTokens,
-      output_tokens: outputTokens,
-    };
-
-    // Add cache_read_input_tokens if present
-    if (cacheReadTokens > 0) {
-      state.usage.cache_read_input_tokens = cacheReadTokens;
-    }
-
-    // Add cache_creation_input_tokens if present
-    if (cacheCreateTokens > 0) {
-      state.usage.cache_creation_input_tokens = cacheCreateTokens;
-    }
-
-    // Note: completion_tokens_details.reasoning_tokens is already included in output_tokens
-    // No need to add separately as Claude expects total output_tokens
+  // Add cache_read_input_tokens if present
+  if (cacheReadTokens > 0) {
+    state.usage.cache_read_input_tokens = cacheReadTokens;
   }
 
+  // Add cache_creation_input_tokens if present
+  if (cacheCreateTokens > 0) {
+    state.usage.cache_creation_input_tokens = cacheCreateTokens;
+  }
+
+  // Note: completion_tokens_details.reasoning_tokens is already included in output_tokens
+  // No need to add separately as Claude expects total output_tokens
+}
+
+// Convert OpenAI stream chunk to Claude format
+export function openaiToClaudeResponse(chunk, state) {
+  if (!chunk && !state.pendingClaudeFinishChoice) return null;
+
+  const results = [];
+  const chunkUsage = chunk?.usage;
+  const hasChunkUsage = chunkUsage && typeof chunkUsage === "object";
+
+  // Usage must be harvested BEFORE the choices guard: many OpenAI-compatible
+  // upstreams (Fireworks, vLLM, Together, …) deliver the authoritative usage
+  // block — including prompt_tokens_details.cached_tokens — on a trailing
+  // usage-only chunk shaped `{"choices":[],"usage":{...}}`. Returning early on
+  // that chunk discarded the real numbers and left downstream accounting on
+  // OmniRoute's own tokenizer estimate (#11817).
+  //
+  // Harvesting alone is not enough: if the finish_reason chunk arrives BEFORE
+  // this trailing usage chunk (the normal order for these upstreams), the
+  // finish block below fires immediately and emits message_delta with
+  // whatever state.usage held at that moment — zero/stale, since the real
+  // trailing chunk hasn't been seen yet. The finish deferral below
+  // (pendingClaudeFinishChoice) holds the terminal emission open until either
+  // real usage has arrived or a genuine flush forces it, so the message_delta
+  // actually sent to the client carries the correct numbers (#11817 follow-up).
+  if (chunk) trackUsageFromChunk(chunk, state);
+
+  const chunkChoice = chunk?.choices?.[0];
+  const flushingPendingFinish = !chunkChoice && Boolean(state.pendingClaudeFinishChoice);
+  const choice = chunkChoice || state.pendingClaudeFinishChoice;
+  if (!choice) return null;
+  if (flushingPendingFinish) state.pendingClaudeFinishChoice = null;
+  const delta = choice.delta;
   // First chunk - ALWAYS send message_start first
   if (!state.messageStartSent) {
     state.messageStartSent = true;
@@ -489,6 +514,11 @@ export function openaiToClaudeResponse(chunk, state) {
   // guard therefore misfired and silently dropped the terminal message_delta/message_stop
   // for Responses→Claude streams (#5828 regression).
   if (choice.finish_reason && !state.claudeFinishEmitted) {
+    if (!hasChunkUsage && !flushingPendingFinish) {
+      state.pendingClaudeFinishChoice = choice;
+      return results.length > 0 ? results : null;
+    }
+
     state.claudeFinishEmitted = true;
     stopThinkingBlock(state, results);
     stopTextBlock(state, results);

scripts/check/check-workflows.mjs

@@ -42,6 +42,7 @@ import { execFileSync, spawnSync } from "node:child_process";
 import fs from "node:fs";
 import path from "node:path";
 import { pathToFileURL } from "node:url";
+import { findProvenanceOnSelfHosted, formatProvenanceFinding } from "./lib/provenanceRunner.mjs";
 
 const ROOT = process.cwd();
 const WORKFLOWS_DIR = path.join(ROOT, ".github", "workflows");
@@ -275,6 +276,23 @@ export function runZizmor(workflowsDir) {
 // Main
 // ---------------------------------------------------------------------------
 
+/**
+ * Hard rule (not a lint count): `--provenance` inside a job that runs on a
+ * self-hosted runner. npm answers 422 at the registry, and in v3.8.50 that
+ * answer only came after the tag, the GitHub Release and the Docker images were
+ * already out. Blocks under --strict AND --ratchet (the CI mode); plain mode
+ * reports it like everything else.
+ * @param {string[]} files absolute workflow paths
+ */
+export function runProvenanceRunnerCheck(files) {
+  const findings = [];
+  for (const file of files) {
+    const text = fs.readFileSync(file, "utf8");
+    findings.push(...findProvenanceOnSelfHosted(text, path.relative(ROOT, file)));
+  }
+  return findings;
+}
+
 function main() {
   const hasActionlint = isBinaryAvailable("actionlint");
   const hasZizmor = isBinaryAvailable("zizmor");
@@ -350,6 +368,16 @@ function main() {
     }
   }
 
+  const provenanceFindings = runProvenanceRunnerCheck(workflowFiles);
+  if (provenanceFindings.length > 0) {
+    console.error(
+      `[check-workflows] provenance×self-hosted: ${provenanceFindings.length} finding(s) — HARD RULE:`
+    );
+    provenanceFindings.forEach((f) => console.error(`  ${formatProvenanceFinding(f)}`));
+  } else if (!QUIET) {
+    console.log("[check-workflows] provenance×self-hosted: OK (0 findings)");
+  }
+
   const total = actionlintCount + zizmorCount;
   process.stdout.write(`workflowFindings=${total}\n`);
   process.stdout.write(`actionlintFindings=${actionlintCount}\n`);
@@ -357,6 +385,15 @@ function main() {
   // Read this line with the count above: a finding total is only reproducible against the
   // version that produced it. See zizmorVersion().
   process.stdout.write(`zizmorVersion=${hasZizmor ? zizmorVersion() : "absent"}\n`);
+  process.stdout.write(`provenanceRunnerFindings=${provenanceFindings.length}\n`);
+  if ((STRICT || RATCHET) && provenanceFindings.length > 0) {
+    console.error(
+      `\n[check-workflows] FAIL — ${provenanceFindings.length} job(s) publish with --provenance from a self-hosted runner.\n` +
+        "  npm rejects that with 422 at the registry. Move the upload step to a github-hosted job\n" +
+        "  (see .github/workflows/npm-publish.yml `stage-npm` for the pattern)."
+    );
+    process.exit(1);
+  }
 
   if (STRICT && total > 0) {
     console.error(`\n[check-workflows] FAIL — ${total} workflow finding(s) total (--strict mode).`);

scripts/check/lib/configExpiry.mjs

@@ -0,0 +1,90 @@
+/**
+ * scripts/check/lib/configExpiry.mjs
+ *
+ * Finds dated validity fields in JSON config packs so a test can fail BEFORE
+ * they lapse. Origin: config/alibaba-free-tier-allowlist.json carried
+ * `"validUntil": "2026-08-27"`; on 2026-08-28 the loader started (correctly)
+ * rejecting the pack and a test that asserted "the shipped pack loads" turned
+ * every PR and main red with no commit involved (#11866). A time bomb, not a
+ * regression — and the only kind of defect a diff review can never catch.
+ *
+ * Pure helpers; the repo-wide assertion lives in
+ * tests/unit/config-expiry-time-bomb.test.ts.
+ */
+import fs from "node:fs";
+import path from "node:path";
+
+export const EXPIRY_KEY =
+  /^(validUntil|valid_until|validTo|valid_to|expiresAt|expires_at|expiry|expires)$/;
+const DAY_MS = 86_400_000;
+
+/**
+ * Walks a parsed JSON value and returns every string-valued expiry field.
+ * @returns {{ file: string, keyPath: string, raw: string, expiresAt: number|null }[]}
+ */
+export function collectExpiryFields(value, file, keyPath = []) {
+  const out = [];
+  if (Array.isArray(value)) {
+    value.forEach((v, i) => out.push(...collectExpiryFields(v, file, [...keyPath, String(i)])));
+    return out;
+  }
+  if (!value || typeof value !== "object") return out;
+  for (const [key, v] of Object.entries(value)) {
+    const kp = [...keyPath, key];
+    if (EXPIRY_KEY.test(key) && typeof v === "string") {
+      const ms = Date.parse(v);
+      out.push({ file, keyPath: kp.join("."), raw: v, expiresAt: Number.isFinite(ms) ? ms : null });
+    } else if (v && typeof v === "object") {
+      out.push(...collectExpiryFields(v, file, kp));
+    }
+  }
+  return out;
+}
+
+/** @returns {"expired"|"expiring"|"ok"|"unparseable"} */
+export function classifyExpiry(field, nowMs, warnDays = 7) {
+  if (field.expiresAt === null) return "unparseable";
+  if (field.expiresAt < nowMs) return "expired";
+  if (field.expiresAt < nowMs + warnDays * DAY_MS) return "expiring";
+  return "ok";
+}
+
+/** All *.json under dir, recursively, skipping node_modules. Sorted for stable output. */
+export function walkJsonFiles(dir) {
+  const out = [];
+  const stack = [dir];
+  while (stack.length > 0) {
+    const current = stack.pop();
+    let entries;
+    try {
+      entries = fs.readdirSync(current, { withFileTypes: true });
+    } catch {
+      continue;
+    }
+    for (const e of entries) {
+      const full = path.join(current, e.name);
+      if (e.isDirectory()) {
+        if (e.name !== "node_modules") stack.push(full);
+      } else if (e.isFile() && e.name.endsWith(".json")) {
+        out.push(full);
+      }
+    }
+  }
+  return out.sort();
+}
+
+/**
+ * Scans every JSON file under `dir`; `file` in the result is relative to `dir`
+ * with forward slashes, so allowlists can key on it portably.
+ */
+export function scanConfigExpiry(dir) {
+  return walkJsonFiles(dir).flatMap((f) => {
+    let parsed;
+    try {
+      parsed = JSON.parse(fs.readFileSync(f, "utf8"));
+    } catch {
+      return []; // not this scanner's job to validate JSON
+    }
+    return collectExpiryFields(parsed, path.relative(dir, f).split(path.sep).join("/"));
+  });
+}

scripts/check/lib/provenanceRunner.mjs

@@ -0,0 +1,83 @@
+/**
+ * scripts/check/lib/provenanceRunner.mjs
+ *
+ * npm refuses `--provenance` from a self-hosted runner:
+ *
+ *   422 Unprocessable Entity - Error verifying sigstore provenance bundle:
+ *   Unsupported GitHub Actions runner environment: "self-hosted".
+ *   Only "github-hosted" runners are supported when publishing with provenance.
+ *
+ * v3.8.50 hit this at the very end of a 76-minute publish job — after the tag,
+ * the GitHub Release and the Docker images were already public — because
+ * `USE_VPS_RUNNER` had been turned on (2026-08-02) with no release in between to
+ * surface it. The combination is greppable, so it must fail in CI the moment a
+ * workflow introduces it, not four weeks later at the registry.
+ *
+ * Pure: takes workflow YAML text, returns the offending (job, step) pairs.
+ */
+import { load as yamlLoad } from "js-yaml";
+
+const SELF_HOSTED = /\bself-hosted\b/;
+const EXPRESSION = /\$\{\{/;
+// Lookahead, not \b: `--provenance-file=…` is a different flag (a pre-built
+// bundle) and must not match — a word boundary sits between "e" and "-".
+const PROVENANCE = /(^|\s)--provenance(?=\s|=|$)/m;
+
+/**
+ * Classifies a job's `runs-on` value.
+ * @returns {"self-hosted"|"hosted"|"unknown"}
+ *   "unknown" = an expression with no literal `self-hosted` in it (e.g.
+ *   `${{ matrix.os }}`); the check does not guess, it skips.
+ */
+export function classifyRunsOn(runsOn) {
+  if (runsOn == null) return "unknown";
+  if (typeof runsOn === "string") {
+    if (SELF_HOSTED.test(runsOn)) return "self-hosted";
+    return EXPRESSION.test(runsOn) ? "unknown" : "hosted";
+  }
+  if (Array.isArray(runsOn)) {
+    return runsOn.some((v) => typeof v === "string" && SELF_HOSTED.test(v))
+      ? "self-hosted"
+      : "hosted";
+  }
+  if (typeof runsOn === "object") {
+    // { group: ..., labels: ... } form
+    const labels = runsOn.labels;
+    return classifyRunsOn(Array.isArray(labels) ? labels : labels == null ? "" : String(labels));
+  }
+  return "unknown";
+}
+
+/**
+ * @param {string} yamlText
+ * @param {string} fileName  used only for reporting
+ * @returns {{ file: string, job: string, step: string }[]}
+ */
+export function findProvenanceOnSelfHosted(yamlText, fileName = "<workflow>") {
+  let doc;
+  try {
+    doc = yamlLoad(yamlText);
+  } catch {
+    // actionlint owns syntax; an unparseable file is not this rule's finding.
+    return [];
+  }
+  const jobs =
+    doc && typeof doc === "object" && doc.jobs && typeof doc.jobs === "object" ? doc.jobs : {};
+  const findings = [];
+  for (const [jobName, job] of Object.entries(jobs)) {
+    if (!job || typeof job !== "object") continue;
+    if (classifyRunsOn(job["runs-on"]) !== "self-hosted") continue;
+    const steps = Array.isArray(job.steps) ? job.steps : [];
+    steps.forEach((step, i) => {
+      if (step && typeof step.run === "string" && PROVENANCE.test(step.run)) {
+        findings.push({ file: fileName, job: jobName, step: step.name || `#${i + 1}` });
+      }
+    });
+  }
+  return findings;
+}
+
+/** Human-readable line per finding, used by the CLI. */
+export function formatProvenanceFinding(f) {
+  return `${f.file}: job "${f.job}", step "${f.step}" runs \`--provenance\` on a self-hosted runner — npm rejects that (422). Move the upload to a github-hosted job.`;
+}

scripts/ops/runner-janitor.sh

@@ -1,53 +1,172 @@
 #!/usr/bin/env bash
-# runner-janitor — self-hosted runner box hygiene (WS3.3, v3.8.49 quality plan).
+# runner-janitor — self-hosted runner box hygiene for the .113 pool.
 #
-# The .113 runner box has recurring failure modes that until now were manual
-# discipline: orphaned tmpfs/work dirs filling the disk, and >4 concurrent
-# runners OOM-killing jobs (16 GB box; incidents on the v3.8.47 release day).
-# Install via cron on the box (see docs/ops/RUNNER_BOX.md):
-#   */30 * * * *  /opt/omniroute-ops/runner-janitor.sh >> /var/log/runner-janitor.log 2>&1
+# Runs from cron every 30 min (see docs/ops/RUNNER_BOX.md). It ACTS on what it
+# can prove is safe and ALERTS on what needs an operator decision. Reads of
+# "is this in use?" and the removal happen in the same command, never in two
+# passes: a check-then-delete with a gap is how a live Build job lost its _work
+# on 2026-08-27.
 #
+# Measured box (2026-08-28): 31 GB RAM, 32 cores, 15 GB swap, /tmp = 12 GB
+# tmpfs (RAM!), 188 GB disk. A single `next-build` peaks at ~14 GB, so two
+# concurrent heavy builds saturate the box and three take it down (06:42Z that
+# day: load 56, two jobs lost). The v3.8.50 postmortem (Parte III) has the numbers.
+#
+# What it does, in order:
+#   1) sweep stale artefacts our tooling leaves behind — tmpfs bases after 3 h
+#      (they hold RAM), disk _work/_temp bases after 24 h; only names we create,
+#      only when no process has them open
+#   2) kill zombie builds: a `next-build` older than ZOMBIE_BUILD_MAX_MIN has no
+#      job attached (a real Build step measures ~26 min). On 2026-08-27 one ran
+#      70 minutes after GitHub had already declared its job lost, eating 3.6 GB
+#      and a full core set. KillMode=mixed on the units covers systemctl
+#      stop/restart; this covers the lost-connection path.
+#   3) prune 48 h-old checkouts under _work of runners whose unit is INACTIVE
+#      (stopped runners cannot be mid-job; active ones are never touched)
+#   4) alert: root disk >= DISK_ALERT_PCT, memory PSI full/avg60 >= threshold,
+#      Runner.Listener count above the ceiling (with a per-project breakdown —
+#      the box also hosts OmniHeuris and OmniMind runners)
+#
+# Usage: runner-janitor.sh [--dry-run] [--help]
 # Exit codes: 0 healthy · 1 attention needed (printed to stdout for the log).
 set -euo pipefail
 
-MAX_ACTIVE_RUNNERS="${MAX_ACTIVE_RUNNERS:-4}"
-DISK_ALERT_PCT="${DISK_ALERT_PCT:-85}"
-WORK_DIR_MAX_AGE_HOURS="${WORK_DIR_MAX_AGE_HOURS:-24}"
-STATUS=0
-
-echo "[janitor] $(date -u +%FT%TZ) start"
-
-# 1) Sweep stale runner temp/work leftovers (>24h — no legitimate job runs that long).
-# Hardened for a root cron on world-writable paths: never follow a symlinked base
-# (a compromised runner could plant one), -P + -xdev so the sweep cannot traverse
-# out of the filesystem, and patterns narrowed to names OUR tooling creates
-# (no generic tmp* — unrelated system temp files are out of scope).
-for base in /tmp /home/*/actions-runner*/_work/_temp; do
-  [ -d "$base" ] || continue
-  [ -L "$base" ] && { echo "[janitor] skip symlinked base: $base"; continue; }
-  find -P "$base" -xdev -maxdepth 1 \( -name 'runner-*' -o -name 'omniroute-*' \) \
-    ! -type l -mmin +$((WORK_DIR_MAX_AGE_HOURS * 60)) -exec rm -rf {} + 2>/dev/null || true
+DRY_RUN=0
+for arg in "$@"; do
+  case "$arg" in
+    --dry-run) DRY_RUN=1 ;;
+    -h|--help)
+      sed -n '2,32p' "$0" | sed 's/^# \{0,1\}//'
+      exit 0 ;;
+    *) echo "unknown argument: $arg" >&2; exit 2 ;;
+  esac
 done
-echo "[janitor] stale temp sweep done"
 
-# 2) Disk pressure — alert loudly before SQLITE_FULL kills jobs mid-run.
-USAGE=$(df --output=pcent / | tail -1 | tr -dc '0-9')
-if [ "$USAGE" -ge "$DISK_ALERT_PCT" ]; then
-  echo "[janitor] ⚠ ROOT DISK ${USAGE}% >= ${DISK_ALERT_PCT}% — clean before the next heavy run"
-  STATUS=1
+MAX_ACTIVE_RUNNERS="${MAX_ACTIVE_RUNNERS:-8}"
+DISK_ALERT_PCT="${DISK_ALERT_PCT:-85}"
+TMPFS_MAX_AGE_HOURS="${TMPFS_MAX_AGE_HOURS:-3}"
+WORK_TEMP_MAX_AGE_HOURS="${WORK_TEMP_MAX_AGE_HOURS:-24}"
+WORK_CHECKOUT_MAX_AGE_HOURS="${WORK_CHECKOUT_MAX_AGE_HOURS:-48}"
+ZOMBIE_BUILD_MAX_MIN="${ZOMBIE_BUILD_MAX_MIN:-75}"
+ZOMBIE_BUILD_COMM="${ZOMBIE_BUILD_COMM:-next-build}"
+PSI_FULL_AVG60_ALERT="${PSI_FULL_AVG60_ALERT:-10}"
+# Overridable so the unit test can point everything at a fixture tree.
+JANITOR_TMP_BASES="${JANITOR_TMP_BASES-/tmp}"
+JANITOR_WORK_TEMP_BASES="${JANITOR_WORK_TEMP_BASES-/opt/actions-runner*/_work/_temp /home/*/actions-runner*/_work/_temp}"
+JANITOR_RUNNER_DIRS="${JANITOR_RUNNER_DIRS-/opt/actions-runner*}"
+JANITOR_PSI_FILE="${JANITOR_PSI_FILE:-/proc/pressure/memory}"
+JANITOR_DF_PATH="${JANITOR_DF_PATH:-/}"
+
+STATUS=0
+say() { echo "[janitor] $*"; }
+
+# "Is anything using this?" — ONE snapshot of every open path on the box
+# (lsof -Fn), then a prefix match per candidate. `lsof +D <dir>` walks the whole
+# tree instead and took minutes on a 5 GB leftover — unusable from cron. An
+# absent lsof means "cannot prove idle": the sweep keeps the path and says so.
+LSOF_BIN="${JANITOR_LSOF:-lsof}"
+have_busy_tools() { command -v "$LSOF_BIN" >/dev/null 2>&1; }
+SNAP=""
+cleanup() { [ -n "$SNAP" ] && rm -f -- "$SNAP"; }
+trap cleanup EXIT
+# One lsof for the whole run (~13 s / 83k lines on the box), kept ONLY for the
+# bases we sweep — 460 candidates grepping a re-printed 83k-line string was the
+# slow part, not lsof itself.
+snapshot_open_paths() {
+  have_busy_tools || return 0
+  SNAP=$(mktemp) || return 0
+  local prefixes="" b
+  for b in $JANITOR_TMP_BASES $JANITOR_WORK_TEMP_BASES; do [ -d "$b" ] && prefixes="$prefixes"$'\n'"$b/"; done
+  # -F n: one "n<path>" line per open file; -w: no warnings
+  "$LSOF_BIN" -w -Fn 2>/dev/null | sed -n 's/^n//p' | grep -F -f <(printf '%s' "$prefixes" | sed '/^$/d') > "$SNAP" 2>/dev/null || true
+}
+is_busy() {
+  local p="$1"
+  [ -n "$SNAP" ] && [ -s "$SNAP" ] || return 1
+  # exact path, or anything beneath it when it is a directory
+  grep -qxF -- "$p" "$SNAP" && return 0
+  [ -d "$p" ] && grep -qF -- "$p/" "$SNAP"
+}
+
+# sweep <base> <max-age-minutes>: only names our tooling creates, never through
+# a symlinked base, never across a filesystem, and remove+check in one step.
+sweep() {
+  local base="$1" max_min="$2" p
+  [ -d "$base" ] || return 0
+  [ -L "$base" ] && { say "skip symlinked base: $base"; return 0; }
+  while IFS= read -r -d '' p; do
+    if ! have_busy_tools; then say "cannot prove idle (lsof missing — apt install lsof), kept: $p"; STATUS=1; continue; fi
+    if is_busy "$p"; then say "busy, kept: $p"; continue; fi
+    if [ "$DRY_RUN" -eq 1 ]; then say "would remove ($(( max_min / 60 ))h+): $p"; else rm -rf -- "$p" && say "removed ($(( max_min / 60 ))h+): $p"; fi
+  done < <(find -P "$base" -xdev -mindepth 1 -maxdepth 1 \
+      \( -name 'runner-*' -o -name 'omniroute-*' -o -name 'next-build*' -o -name 'e2e-build.tar.gz' \) \
+      ! -type l -mmin "+$max_min" -print0 2>/dev/null || true)
+}
+
+say "$(date -u +%FT%TZ) start${DRY_RUN:+ (dry-run=$DRY_RUN)} busy-tools=$(have_busy_tools && echo ok || echo MISSING)"
+
+# 1) stale artefacts — tmpfs is RAM, so it gets the short fuse
+snapshot_open_paths
+for base in $JANITOR_TMP_BASES; do sweep "$base" $(( TMPFS_MAX_AGE_HOURS * 60 )); done
+for base in $JANITOR_WORK_TEMP_BASES; do sweep "$base" $(( WORK_TEMP_MAX_AGE_HOURS * 60 )); done
+say "stale temp sweep done"
+
+# 2) zombie builds
+ZOMBIES=0
+while read -r pid etimes comm; do
+  [ -n "${pid:-}" ] || continue
+  if [ "$etimes" -gt $(( ZOMBIE_BUILD_MAX_MIN * 60 )) ]; then
+    say "⚠ zombie build pid=$pid comm=$comm age=$(( etimes / 60 ))min > ${ZOMBIE_BUILD_MAX_MIN}min — no job runs this long"
+    if [ "$DRY_RUN" -eq 1 ]; then say "[dry-run] would: kill -TERM $pid (then -KILL)"; else
+      kill -TERM "$pid" 2>/dev/null || true; sleep 10
+      kill -0 "$pid" 2>/dev/null && { kill -KILL "$pid" 2>/dev/null || true; say "  needed SIGKILL"; }
+    fi
+    ZOMBIES=$(( ZOMBIES + 1 )); STATUS=1
+  fi
+done < <(ps -eo pid=,etimes=,comm= 2>/dev/null | awk -v c="$ZOMBIE_BUILD_COMM" '$3 ~ ("^" c) {print $1, $2, $3}' || true)
+say "zombie builds: $ZOMBIES"
+
+# 3) old checkouts of STOPPED runners
+for d in $JANITOR_RUNNER_DIRS; do
+  [ -d "$d" ] && [ -f "$d/.runner" ] || continue
+  agent=$(grep -o '"agentName": *"[^"]*"' "$d/.runner" 2>/dev/null | sed 's/.*"\([^"]*\)"$/\1/')
+  [ -n "$agent" ] || continue
+  unit=$(systemctl list-units --plain --no-legend "actions.runner.*.${agent}.service" 2>/dev/null | awk 'NR==1{print $1}')
+  [ -n "$unit" ] || continue
+  if systemctl is-active --quiet "$unit"; then continue; fi
+  while IFS= read -r -d '' co; do
+    if [ "$DRY_RUN" -eq 1 ]; then say "would prune checkout of stopped runner $agent: $co"; else rm -rf -- "$co" && say "pruned checkout of stopped runner $agent: $co"; fi
+  done < <(find -P "$d/_work" -xdev -mindepth 2 -maxdepth 2 -type d -mmin "+$(( WORK_CHECKOUT_MAX_AGE_HOURS * 60 ))" -print0 2>/dev/null || true)
+done
+
+# 4a) disk
+USAGE=$(df --output=pcent "$JANITOR_DF_PATH" 2>/dev/null | tail -1 | tr -dc '0-9')
+if [ "${USAGE:-0}" -ge "$DISK_ALERT_PCT" ]; then
+  say "⚠ ROOT DISK ${USAGE}% >= ${DISK_ALERT_PCT}% — clean before the next heavy run"; STATUS=1
 else
-  echo "[janitor] disk ${USAGE}% OK"
+  say "disk ${USAGE:-?}% OK"
 fi
 
-# 3) Concurrency ceiling — 8-wide OOMed the 16 GB box twice on release day;
-#    4 is the proven ceiling. This CODIFIES the rule that was manual discipline.
+# 4b) memory pressure (PSI) — the box swapped its way through the v3.8.50 publish
+if [ -r "$JANITOR_PSI_FILE" ]; then
+  FULL60=$(awk '/^full/ {for(i=1;i<=NF;i++) if ($i ~ /^avg60=/) {sub("avg60=","",$i); print $i}}' "$JANITOR_PSI_FILE" 2>/dev/null || echo "")
+  if [ -n "$FULL60" ] && awk -v v="$FULL60" -v t="$PSI_FULL_AVG60_ALERT" 'BEGIN{exit !(v+0 >= t+0)}'; then
+    say "⚠ MEMORY PRESSURE psi full/avg60=${FULL60}% >= ${PSI_FULL_AVG60_ALERT}% — too many heavy jobs at once"; STATUS=1
+  else
+    say "memory psi full/avg60=${FULL60:-n/a}% OK"
+  fi
+fi
+
+# 4c) concurrency ceiling — alert with a breakdown; the fix is fewer/labelled
+#     runners (an operator decision), not killing listeners from cron.
 ACTIVE=$(pgrep -fc "Runner.Listener" || true)
+OMNI=$(pgrep -fc "actions-runner-omniroute[^ ]*/bin[^ ]*/Runner.Listener" || true)
 if [ "${ACTIVE:-0}" -gt "$MAX_ACTIVE_RUNNERS" ]; then
-  echo "[janitor] ⚠ ${ACTIVE} Runner.Listener processes > ceiling ${MAX_ACTIVE_RUNNERS} — stop the extra runners (systemctl stop actions.runner.<name>)"
+  say "⚠ ${ACTIVE} Runner.Listener processes (omniroute=${OMNI:-0}, other=$(( ${ACTIVE:-0} - ${OMNI:-0} ))) > ceiling ${MAX_ACTIVE_RUNNERS} — stop idle extras: systemctl stop <unit> only when it has no Runner.Worker child"
   STATUS=1
 else
-  echo "[janitor] runners active: ${ACTIVE:-0}/${MAX_ACTIVE_RUNNERS} OK"
+  say "runners active: ${ACTIVE:-0}/${MAX_ACTIVE_RUNNERS} (omniroute=${OMNI:-0}) OK"
 fi
 
-echo "[janitor] done status=$STATUS"
+say "done status=$STATUS"
 exit "$STATUS"

src/app/(dashboard)/dashboard/combos/page.tsx

@@ -2037,7 +2037,9 @@ function ComboFormModal({ isOpen, combo, onClose, onSave, activeProviders, combo
   const [showAdvanced, setShowAdvanced] = useState(false);
   const [config, setConfig] = useState(sanitizeComboRuntimeConfig(combo?.config));
   // Validate persisted enum; ensure reset on combo change not just first mount.
-  const initialSortMethod = normalizeSortMethod(config.modelSort?.method);
+  const initialSortMethod = normalizeSortMethod(
+    (config.modelSort as { method?: unknown } | undefined)?.method
+  );
   const [sortMethod, setSortMethod] = useState<SortMethod>(initialSortMethod);
   useEffect(() => {
     // Sync point: when the combo identity changes, re-derive sort method.
@@ -2733,13 +2735,15 @@ function ComboFormModal({ isOpen, combo, onClose, onSave, activeProviders, combo
         const rankings = await fetchProviderRankings();
         // Functional note: `next` is the post-batch snapshot. Concurrent single-add
         // racing this batch is low-probability single-user; last write wins.
-        const sorted = await sortComboStepsByScore(next, rankings);
-        setModels(sorted);
+        const sorted = await sortComboStepsByScore(next as ComboStep[], rankings);
+        setModels(sorted as typeof next);
       } catch {
         setModels(next);
       }
     } else {
-      setModels(sortComboStepsSync(next, currentMethod as "provider" | "name"));
+      setModels(
+        sortComboStepsSync(next as ComboStep[], currentMethod as "provider" | "name") as typeof next
+      );
     }
     setBuilderError("");
   };
@@ -3642,7 +3646,11 @@ function ComboFormModal({ isOpen, combo, onClose, onSave, activeProviders, combo
               </div>
 
               <div className="flex items-center justify-between gap-2 mb-2">
-                <ComboSortSelect value={sortMethod} onChange={handleSortChange} t={t} />
+                <ComboSortSelect
+                  value={sortMethod}
+                  onChange={handleSortChange}
+                  t={(k, f) => getI18nOrFallback(t, k, f)}
+                />
               </div>
 
               {models.length === 0 ? (

src/app/(dashboard)/dashboard/providers/[id]/components/ProviderPageHeader.tsx

@@ -12,7 +12,7 @@ interface ProviderInfo {
   id: string;
   name: string;
   website?: string;
-  color: string;
+  color?: string;
   apiType?: string;
   /** Optional operator-supplied remote icon URL (#2166) for compatible provider nodes. */
   iconUrl?: string;

src/app/api/playground/simulate-route/route.ts

@@ -153,32 +153,60 @@ export async function POST(request: Request) {
         return NextResponse.json({ error: "Combo not found" }, { status: 404 });
       }
       const persistedSteps = Array.isArray(combo.models) ? combo.models : [];
-      const modelSteps = persistedSteps.filter(
-        (step: any) =>
+      let unsupportedStepCount = 0;
+      const targets = persistedSteps.flatMap((step: any) => {
+        if (
           typeof step === "string" ||
           ((step.kind === undefined || step.kind === "model") && typeof step.model === "string")
-      );
-      const unsupportedStepCount = persistedSteps.length - modelSteps.length;
+        ) {
+          const value = typeof step === "string" ? step : step.model;
+          const separator = value.indexOf("/");
+          const parsedProvider = separator === -1 ? undefined : value.slice(0, separator);
+          const parsedModel = separator === -1 ? value : value.slice(separator + 1);
+
+          return [
+            {
+              provider:
+                typeof step === "string"
+                  ? parsedProvider || "unknown"
+                  : step.providerId || step.provider || parsedProvider || "unknown",
+              model: parsedModel,
+              weight: typeof step === "string" ? undefined : step.weight,
+            },
+          ];
+        }
+
+        // #11822 follow-up (see #11882): surface combo-ref and provider-wildcard
+        // steps with a specific warning instead of folding them into a generic
+        // "unsupported step" count. Structural combo references and wildcard
+        // expansion are out of scope for this route-local simulator.
+        if (step?.kind === "combo-ref") {
+          warnings.push(
+            `Step references combo "${String(step.comboName)}" — nested combos are not expanded by the simulator.`
+          );
+          return [];
+        }
+        if (step?.kind === "provider-wildcard") {
+          warnings.push(
+            `Step "${String(step.providerId)}/${String(step.modelPattern)}" is a provider wildcard — expanded at runtime, shown here unresolved.`
+          );
+          return [
+            {
+              provider: String(step.providerId ?? "unknown"),
+              model: String(step.modelPattern ?? "*"),
+              weight: typeof step.weight === "number" ? step.weight : undefined,
+            },
+          ];
+        }
+
+        unsupportedStepCount += 1;
+        return [];
+      });
       if (unsupportedStepCount > 0) {
         warnings.push(
           `Skipped ${unsupportedStepCount} unsupported persisted combo ${unsupportedStepCount === 1 ? "step" : "steps"}.`
         );
       }
-      const targets = modelSteps.map((step: any) => {
-        const value = typeof step === "string" ? step : step.model;
-        const separator = value.indexOf("/");
-        const parsedProvider = separator === -1 ? undefined : value.slice(0, separator);
-        const parsedModel = separator === -1 ? value : value.slice(separator + 1);
-
-        return {
-          provider:
-            typeof step === "string"
-              ? parsedProvider || "unknown"
-              : step.providerId || step.provider || parsedProvider || "unknown",
-          model: parsedModel,
-          weight: typeof step === "string" ? undefined : step.weight,
-        };
-      });
       comboInfo = { name: combo.name, strategy: combo.strategy, targets };
     } else if (body.combo) {
       comboInfo = body.combo;

src/app/api/v1/models/catalog.ts

@@ -31,6 +31,7 @@ import { getAllMusicModels } from "@omniroute/open-sse/config/musicRegistry";
 import {
   getRegistryModelThinkingEfforts,
   getRegistryThinkingEfforts,
+  providerUsesAuthoritativeLiveCatalog,
   REGISTRY,
 } from "@omniroute/open-sse/config/providerRegistry";
 import { CODEX_NATIVE_UNPREFIXED_MODELS } from "@omniroute/open-sse/services/model";
@@ -994,11 +995,13 @@ async function buildUnifiedModelsResponseCore(
         // the fix, a provider with any synced model silently dropped ALL its
         // static models.
         //
-        // Cursor exclusive listing: when an active synced catalog exists, drop
-        // ALL static rows (including effort variants) so Test All / clients only
-        // see live AvailableModels + injected auto*.
+        // An authoritative active synced catalog replaces the static registry.
+        // Partial discovery providers still use exact-id coverage suppression so
+        // their intentionally omitted static routes remain available.
         const syncedForProvider = syncedModelIdsByCanonicalProvider.get(canonicalProviderId);
-        const exclusiveListing = providerUsesExclusiveSyncedListing(canonicalProviderId);
+        const exclusiveListing =
+          providerUsesExclusiveSyncedListing(canonicalProviderId) ||
+          providerUsesAuthoritativeLiveCatalog(canonicalProviderId);
         const providerHasSynced = syncedForProvider !== undefined && syncedForProvider.size > 0;
         const coveredBySynced = shouldSuppressStaticModelForExclusiveListing({
           exclusiveListing,
@@ -1227,7 +1230,7 @@ async function buildUnifiedModelsResponseCore(
             continue;
           }
 
-          if (includeAlias) {
+          if (includeAlias || Boolean(prefix)) {
             models.push({
               id: aliasId,
               object: "model",
@@ -1239,7 +1242,7 @@ async function buildUnifiedModelsResponseCore(
               ...syncedFields,
             });
           }
-          if (includeAlias && modelType === "audio") {
+          if ((includeAlias || Boolean(prefix)) && modelType === "audio") {
             models.push({
               id: aliasId,
               object: "model",
@@ -1652,7 +1655,7 @@ async function buildUnifiedModelsResponseCore(
             ? getCustomVisionCapabilityFields(model, aliasId, modelId)
             : null;
 
-          if (includeAlias) {
+          if (includeAlias || Boolean(prefix)) {
             models.push({
               id: aliasId,
               object: "model",
@@ -1770,7 +1773,7 @@ async function buildUnifiedModelsResponseCore(
         const visionFields =
           getVisionCapabilityFields(aliasId) || getVisionCapabilityFields(modelId);
 
-        if (includeAlias) {
+        if (includeAlias || Boolean(nodePrefix)) {
           models.push({
             id: aliasId,
             object: "model",

src/i18n/messages/pt-BR.json

@@ -7679,7 +7679,8 @@
     "backupCleanupSuccess": "Excluído(s) {backups} conjunto(s) de backup e {files} arquivo(s).",
     "backupCleanupFailed": "Falha ao limpar backups do banco de dados",
     "purgeQuotaSnapshotsSuccess": "{count} snapshots de cota removidos",
-    "purgeQuotaSnapshotsFailed": "Falha ao remover snapshots de cota",    "purgeCallLogsSuccess": "{count} logs de chamadas removidos",
+    "purgeQuotaSnapshotsFailed": "Falha ao remover snapshots de cota",
+    "purgeCallLogsSuccess": "{count} logs de chamadas removidos",
     "purgeCallLogsFailed": "Falha ao remover logs de chamadas",
     "purgeDetailedLogsSuccess": "{count} logs detalhados removidos",
     "purgeDetailedLogsFailed": "Falha ao remover logs detalhados",
@@ -7689,7 +7690,8 @@
     "invalidJsonFileType": "Tipo de arquivo inválido. Apenas arquivos .json são permitidos.",
     "legacyJsonImportSuccess": "JSON legado importado com sucesso!",
     "jsonImportFailed": "Falha ao importar JSON",
-    "jsonImportError": "Erro durante a importação de JSON",    "storagePurgeData": "Limpar dados",
+    "jsonImportError": "Erro durante a importação de JSON",
+    "storagePurgeData": "Limpar dados",
     "storagePurgeDataDesc": "Excluir imediatamente todos os registros sem aplicar verificações de retenção. Use com cautela.",
     "storageRetentionCleanup": "Configurações de Retenção",
     "storageRetentionCleanupDesc": "Configure a retenção de registros operacionais e a limpeza de backup do banco de dados.",
@@ -13937,5 +13939,17 @@
     "cta": "Obter uma chave de API",
     "partnerLinkNote": "Link de parceiro",
     "dismissAriaLabel": "Dispensar"
+  },
+  "combo": {
+    "sort": {
+      "label": "Ordenar por",
+      "method": {
+        "manual": "Manual",
+        "provider": "Provedor",
+        "score": "Pontuação (modelos gratuitos)",
+        "name": "Nome"
+      },
+      "scoreHint": "A ordenação por pontuação vale só para provedores gratuitos; os demais ficam onde estão."
+    }
   }
 }

src/i18n/messages/vi.json

@@ -13939,5 +13939,17 @@
     "cta": "Lấy khóa API",
     "partnerLinkNote": "Liên kết đối tác",
     "dismissAriaLabel": "Đóng"
+  },
+  "combo": {
+    "sort": {
+      "label": "Sắp xếp theo",
+      "method": {
+        "manual": "Thủ công",
+        "provider": "Nhà cung cấp",
+        "score": "Điểm (mô hình miễn phí)",
+        "name": "Tên"
+      },
+      "scoreHint": "Xếp hạng theo điểm chỉ áp dụng cho nhà cung cấp miễn phí; các nhà cung cấp khác giữ nguyên vị trí."
+    }
   }
 }

src/lib/providers/xai/thinking.ts

@@ -12,14 +12,16 @@
  *   - honor explicit caller intent verbatim
  */
 
-const VALID_EFFORTS = new Set(["minimal", "low", "medium", "high"]);
+const VALID_EFFORTS = new Set(["minimal", "low", "medium", "high", "xhigh"]);
 
-export type ReasoningEffort = "minimal" | "low" | "medium" | "high";
+export type ReasoningEffort = "minimal" | "low" | "medium" | "high" | "xhigh";
 
 export function normalizeXaiReasoningEffort(effort: unknown): ReasoningEffort | undefined {
   if (typeof effort !== "string") return undefined;
   const normalized = effort.toLowerCase();
-  if (normalized === "max" || normalized === "xhigh") return "high";
+  // "max" is not an xAI tier; "xhigh" is real on grok-4.6+ and xAI itself
+  // degrades it to "high" on older models, so passing it through is always safe.
+  if (normalized === "max") return "high";
   return VALID_EFFORTS.has(normalized) ? (normalized as ReasoningEffort) : undefined;
 }
 

src/lib/skills/injection.ts

@@ -63,9 +63,17 @@ function normalizeInputSchema(input: Record<string, unknown>): Record<string, un
   if (typeof input.type === "string") {
     return input;
   }
+  // Expand shorthand values: skills may declare `{ "content": "string" }`
+  // instead of `{ "content": { "type": "string" } }`. Forwarding the shorthand
+  // verbatim produces invalid JSON Schema, which strict-validating upstreams
+  // (Zhipu GLM behind Console Go) reject with a 400 for the entire request.
+  const properties: Record<string, unknown> = {};
+  for (const [key, value] of Object.entries(input)) {
+    properties[key] = typeof value === "string" ? { type: value } : value;
+  }
   return {
     type: "object",
-    properties: input,
+    properties,
   };
 }
 

src/lib/usage/providerLimits.ts

@@ -13,12 +13,7 @@ import {
 } from "@/lib/db/providerLimits";
 import { syncToCloud } from "@/lib/cloudSync";
 import { setQuotaCache } from "@/domain/quotaCache";
-import {
-  buildClaudeExtraUsageConnectionUpdate,
-  CLAUDE_EXTRA_USAGE_ERROR_SOURCE,
-  isClaudeExtraUsageBlockEnabled,
-  isClaudeExtraUsageQueued,
-} from "@/lib/providers/claudeExtraUsage";
+import { buildClaudeExtraUsageConnectionUpdate } from "@/lib/providers/claudeExtraUsage";
 import { isConnectionUnavailableToAuxiliaryActivity } from "@/lib/exclusiveLeaseIsolation";
 import { clearRecoveredProviderState } from "@/sse/services/auth";
 import { getMachineId } from "@/shared/utils/machine";
@@ -526,48 +521,7 @@ export function shouldClearErrorStateOnValidProbe(
  * — keeps the connection locked, matching the kimi-coding partial-refresh
  * semantics.
  */
-function windowStillExhaustedAfterRealReset(value: unknown, nowMs: number): boolean {
-  if (!isRecord(value)) return false;
-  if (value.unlimited === true) return false;
-  const remaining =
-    typeof value.remaining === "number"
-      ? value.remaining
-      : typeof value.remainingPercentage === "number"
-        ? value.remainingPercentage
-        : null;
-  if (remaining !== null && remaining > 0) return false;
-  if (value.resetAt == null) return true;
-  const resetMs = Date.parse(String(value.resetAt));
-  if (Number.isNaN(resetMs)) return true;
-  return resetMs > nowMs;
-}
 
-function isQuotaExhaustedCooldownReleasable(
-  connection: Pick<
-    ProviderConnectionLike,
-    "lastErrorType" | "lastErrorSource" | "provider" | "providerSpecificData"
-  >,
-  usage: JsonRecord
-): boolean {
-  if (connection.lastErrorType !== "quota_exhausted") return false;
-  // An extra-usage block is a POLICY lock, not a quota window: the session and
-  // weekly windows genuinely look recovered in the very same fetch, so the
-  // window scan below would happily release it. It stays locked while the
-  // policy is on and upstream still reports extra usage queued.
-  if (
-    connection.lastErrorSource === CLAUDE_EXTRA_USAGE_ERROR_SOURCE &&
-    isClaudeExtraUsageBlockEnabled(connection.provider, connection.providerSpecificData) &&
-    isClaudeExtraUsageQueued(usage)
-  ) {
-    return false;
-  }
-  const quotas = usage?.quotas;
-  if (!isRecord(quotas)) return false;
-  const values = Object.values(quotas);
-  if (values.length === 0) return false;
-  const nowMs = Date.now();
-  return !values.some((value) => windowStillExhaustedAfterRealReset(value, nowMs));
-}
 
 /**
  * Is an explicit cooldown still in the future?
@@ -600,17 +554,17 @@ export async function maybeClearRecoveredQuotaState(
   if (!hasUsableQuota(usage)) return connection;
   if (isTerminalStatusForQuotaRecovery(connection.testStatus)) return connection;
   if (hasActiveCooldown(connection)) {
-    // #11355 made an active rateLimitedUntil an unconditional stop, which is right
-    // for an upstream-derived cooldown but over-broad for the one case #10534 was
-    // built for: a Claude-subscription 429 persists a SYNTHETIC 1h cooldown because
-    // the upstream sent no parseable reset. When the later poll shows every window
-    // that governs this connection has really reset WITH quota available, holding
-    // that synthetic cooldown just deadlocks the connection for an hour.
-    //
-    // Narrow by design: only lastErrorType "quota_exhausted" (the synthetic-cooldown
-    // writer) is eligible, and a single still-exhausted or unknown-reset window keeps
-    // the lock. Every other reason keeps #11355/#11277 semantics untouched.
-    if (!isQuotaExhaustedCooldownReleasable(connection, usage)) return connection;
+    // A future rateLimitedUntil written from a real upstream signal is a hard
+    // statement no poller may overrule (#11277) — executor-sourced rate limits
+    // and extra-usage policy blocks included. Only a SYNTHETIC cooldown (a
+    // quota_exhausted lock persisted without an upstream reset, e.g. the
+    // Claude-subscription poller's 1h lockout) yields to positive live-window
+    // evidence that the real quota has already replenished past its reset.
+    const syntheticRecoveryOverride =
+      connection.lastErrorType === "quota_exhausted" &&
+      connection.lastErrorSource !== "extra_usage" &&
+      syntheticCooldownOutlivedByRealWindows(usage);
+    if (!syntheticRecoveryOverride) return connection;
   }
 
   const hasTransientState =

src/shared/components/ProviderIcon.tsx

@@ -335,11 +335,19 @@ const ProviderIcon = memo(function ProviderIcon({
   fallbackColor,
 }: ProviderIconProps) {
   const { isDark } = useTheme();
-  const normalizedId = PROVIDER_ICON_ALIASES[providerId.toLowerCase()] || providerId.toLowerCase();
-  const localSvgId = LOCAL_SVG_ALIASES[normalizedId] || normalizedId;
+  // Own-property guards: a providerId such as "constructor" or "__proto__" otherwise
+  // resolves through Object.prototype, yielding a truthy-looking value that corrupts
+  // downstream lookups instead of falling through to the unknown-provider path (#11853).
+  const providerIdLower = providerId.toLowerCase();
+  const normalizedId = Object.hasOwn(PROVIDER_ICON_ALIASES, providerIdLower)
+    ? PROVIDER_ICON_ALIASES[providerIdLower]
+    : providerIdLower;
+  const localSvgId = Object.hasOwn(LOCAL_SVG_ALIASES, normalizedId)
+    ? LOCAL_SVG_ALIASES[normalizedId]
+    : normalizedId;
   const usesGenericIcon =
     GENERIC_PROVIDER_IDS.has(normalizedId) || GENERIC_PROVIDER_IDS.has(localSvgId);
-  const themedSvg = THEMED_SVGS[normalizedId];
+  const themedSvg = Object.hasOwn(THEMED_SVGS, normalizedId) ? THEMED_SVGS[normalizedId] : undefined;
   const hasSvg = KNOWN_SVGS.has(localSvgId);
 
   const [failedAssets, setFailedAssets] = useState<Record<string, true>>({});

src/shared/components/lobeProviderIcons.ts

@@ -484,9 +484,17 @@ export function getLobeProviderIcon(
   providerId: string,
   type: "mono" | "color" = "color"
 ): LobeIconComponent | null {
-  const iconKey = LOBE_PROVIDER_ALIASES[providerId.toLowerCase()];
-  if (!iconKey) return null;
+  if (typeof providerId !== "string") return null;
+  const aliasKey = providerId.toLowerCase();
+  // Own-property guards: a providerId such as "constructor" or "__proto__"
+  // otherwise resolves through Object.prototype, yielding a truthy iconKey
+  // whose LOBE_ICON_COMPONENTS lookup is undefined -> `entry.color` throws and
+  // takes down the whole providers dashboard via the error boundary.
+  if (!Object.hasOwn(LOBE_PROVIDER_ALIASES, aliasKey)) return null;
+  const iconKey = LOBE_PROVIDER_ALIASES[aliasKey];
+  if (!iconKey || !Object.hasOwn(LOBE_ICON_COMPONENTS, iconKey)) return null;
 
   const entry = LOBE_ICON_COMPONENTS[iconKey];
+  if (!entry) return null;
   return type === "color" && entry.color ? entry.color : entry.mono;
 }

src/shared/constants/providers/apikey/gateways.ts

@@ -695,7 +695,7 @@ export const APIKEY_PROVIDERS_GATEWAYS = {
       "Anonymous access to Groq requires proof-of-work cake credits from g4f.dev/chat; alternatively, use a g4f.dev member API key. Limits vary.",
     passthroughModels: true,
     authHint:
-      "Bake anonymous cake credits at g4f.dev/chat, or create a member API key at g4f.dev/members.html.",
+      "Bake anonymous cake credits at g4f.dev/chat, or use a g4f.dev member key (create one at g4f.dev/members.html).",
     notice: {
       text: "Remote third-party gateway: prompts and request metadata leave OmniRoute and are handled by g4f.space. Its Terms and Privacy links were unavailable when last verified on 2026-08-27.",
       apiKeyUrl: "https://g4f.dev/members.html",
@@ -714,7 +714,7 @@ export const APIKEY_PROVIDERS_GATEWAYS = {
       "Anonymous access to Gemini requires proof-of-work cake credits from g4f.dev/chat; alternatively, use a g4f.dev member API key. Limits vary.",
     passthroughModels: true,
     authHint:
-      "Bake anonymous cake credits at g4f.dev/chat, or create a member API key at g4f.dev/members.html.",
+      "Bake anonymous cake credits at g4f.dev/chat, or use a g4f.dev member key (create one at g4f.dev/members.html).",
     notice: {
       text: "Remote third-party gateway: prompts and request metadata leave OmniRoute and are handled by g4f.space. Its Terms and Privacy links were unavailable when last verified on 2026-08-27.",
       apiKeyUrl: "https://g4f.dev/members.html",
@@ -733,7 +733,7 @@ export const APIKEY_PROVIDERS_GATEWAYS = {
       "Anonymous access to Pollinations requires proof-of-work cake credits from g4f.dev/chat; alternatively, use a g4f.dev member API key. Limits vary.",
     passthroughModels: true,
     authHint:
-      "Bake anonymous cake credits at g4f.dev/chat, or create a member API key at g4f.dev/members.html.",
+      "Bake anonymous cake credits at g4f.dev/chat, or use a g4f.dev member key (create one at g4f.dev/members.html).",
     notice: {
       text: "Remote third-party gateway: prompts and request metadata leave OmniRoute and are handled by g4f.space. Its Terms and Privacy links were unavailable when last verified on 2026-08-27.",
       apiKeyUrl: "https://g4f.dev/members.html",
@@ -752,7 +752,7 @@ export const APIKEY_PROVIDERS_GATEWAYS = {
       "Anonymous access to hosted Ollama requires proof-of-work cake credits from g4f.dev/chat; alternatively, use a g4f.dev member API key. Limits vary.",
     passthroughModels: true,
     authHint:
-      "Bake anonymous cake credits at g4f.dev/chat, or create a member API key at g4f.dev/members.html.",
+      "Bake anonymous cake credits at g4f.dev/chat, or use a g4f.dev member key (create one at g4f.dev/members.html).",
     notice: {
       text: "Remote third-party gateway: prompts and request metadata leave OmniRoute and are handled by g4f.space. Its Terms and Privacy links were unavailable when last verified on 2026-08-27.",
       apiKeyUrl: "https://g4f.dev/members.html",
@@ -771,7 +771,7 @@ export const APIKEY_PROVIDERS_GATEWAYS = {
       "Anonymous access to NVIDIA NIM requires proof-of-work cake credits from g4f.dev/chat; alternatively, use a g4f.dev member API key. Limits vary.",
     passthroughModels: true,
     authHint:
-      "Bake anonymous cake credits at g4f.dev/chat, or create a member API key at g4f.dev/members.html.",
+      "Bake anonymous cake credits at g4f.dev/chat, or use a g4f.dev member key (create one at g4f.dev/members.html).",
     notice: {
       text: "Remote third-party gateway: prompts and request metadata leave OmniRoute and are handled by g4f.space. Its Terms and Privacy links were unavailable when last verified on 2026-08-27.",
       apiKeyUrl: "https://g4f.dev/members.html",

stryker.conf.json

@@ -1,14 +1,14 @@
 {
   "$schema": "https://stryker-mutator.io/schemas/stryker-schema.json",
   "_comment": [
-    "Mutation testing for the ~8 critical modules (Task 11 \u2014 Fase 7).",
-    "NIGHTLY ONLY \u2014 DO NOT run on every PR. Mutation testing is expensive:",
+    "Mutation testing for the ~8 critical modules (Task 11 — Fase 7).",
+    "NIGHTLY ONLY — DO NOT run on every PR. Mutation testing is expensive:",
     "  - Each mutant requires a full test suite execution.",
-    "  - The 8 modules produce ~200\u2013500 mutants; est. 30\u201390 min per run.",
+    "  - The 8 modules produce ~200–500 mutants; est. 30–90 min per run.",
     "  - Wired to the nightly CI workflow (.github/workflows/nightly-mutation.yml),",
     "    NOT to the 'lint' / 'quality-gate' PR jobs.",
     "",
-    "TEST RUNNER \u2014 @stryker-mutator/tap-runner (NOT vitest):",
+    "TEST RUNNER — @stryker-mutator/tap-runner (NOT vitest):",
     "  The 8 critical modules are covered by node:test files in tests/unit/",
     "  (run via `node --import tsx --test`), NOT by vitest. The vitest config",
     "  only includes a small set of .test.tsx + open-sse/**/__tests__ files, so",
@@ -22,24 +22,26 @@
     "  npm install --save-dev @stryker-mutator/core @stryker-mutator/tap-runner",
     "",
     "Run manually:",
-    "  npm run test:mutation                       # full run (slow \u2014 nightly budget)",
+    "  npm run test:mutation                       # full run (slow — nightly budget)",
     "  npx stryker run --dryRunOnly                 # validate the baseline only (no mutants)",
     "  (single-module probe: temporarily narrow `mutate` + `tap.testFiles` in this file)",
     "",
-    "VALIDATED 2026-06-15: `npx stryker run --dryRunOnly` exits 0 \u2014 all 129 covering",
+    "VALIDATED 2026-06-15: `npx stryker run --dryRunOnly` exits 0 — all 129 covering",
     "test files run green in the Stryker sandbox and the perTest coverage map builds for",
     "all 8 instrumented modules (15k+ mutants). The baseline dry-run takes ~20 min with",
     "concurrency=1; the full mutation phase runs on top (advisory, capped by the workflow",
     "timeout). So the nightly produces REAL mutation scores for the 8 modules.",
     "",
-    "Mutation score per module \u2192 quality-baseline.json key 'mutationScore.<module>'",
-    "Direction: up (score can only improve; ratchet blocks drops \u2014 wired in a later INT phase)."
+    "Mutation score per module → quality-baseline.json key 'mutationScore.<module>'",
+    "Direction: up (score can only improve; ratchet blocks drops — wired in a later INT phase)."
   ],
   "packageManager": "npm",
   "incremental": true,
   "incrementalFile": "reports/mutation/stryker-incremental.json",
   "testRunner": "tap",
-  "plugins": ["@stryker-mutator/tap-runner"],
+  "plugins": [
+    "@stryker-mutator/tap-runner"
+  ],
   "tap": {
     "testFiles": [
       "tests/unit/10085-compatible-generic-vs-uuid-credential.test.ts",
@@ -389,7 +391,13 @@
       "tests/unit/vertex-passthrough-model-lockout.test.ts",
       "tests/unit/video-bridge-drilldown-route.test.ts",
       "tests/unit/video-bridge-route-security.test.ts",
-      "tests/unit/xai-agent-tools-passthrough.test.ts"
+      "tests/unit/xai-agent-tools-passthrough.test.ts",
+      "tests/unit/combo/connection-aware-expansion.test.ts",
+      "tests/unit/chatgpt-web-runtime-block.test.ts",
+      "tests/unit/felo-web-runtime-block.test.ts",
+      "tests/unit/microsoft-designer-web-runtime-block.test.ts",
+      "tests/unit/qwen-web-runtime-block.test.ts",
+      "tests/unit/tunnel-routes-error-sanitization.test.ts"
     ],
     "nodeArgs": [
       "--import",
@@ -405,7 +413,7 @@
     ]
   },
   "_mutate_godfiles_excluded_comment": [
-    "2026-06-18 (Onda 2 budget): chatCore.ts + combo.ts \u2014 the two god-files \u2014 were REMOVED",
+    "2026-06-18 (Onda 2 budget): chatCore.ts + combo.ts — the two god-files — were REMOVED",
     "from `mutate`. They dominated ~2/3 of the ~15k mutants; the full 8-module run TIMED OUT",
     "at the 180min nightly cap (run 27705123780: 16:47:33 -> killed 19:47:48 = exactly 180min;",
     "the prior 120min scheduled run also timed out). #4078 made concurrency safe but the",
@@ -425,23 +433,23 @@
     "comboContextCache/idempotency/passthroughHelpers/responseHeaders/sanitization/upstreamTimeouts).",
     "A follow-up then added DEDICATED unit tests for 6 more leaves (tests/unit/chatcore-headers,",
     "-log-truncation, -memory-extraction, -non-streaming-sse, -passthrough-tool-names,",
-    "-executor-helpers \u2014 wired into tap.testFiles above) and added those leaves as batch h",
+    "-executor-helpers — wired into tap.testFiles above) and added those leaves as batch h",
     "(headers/logTruncation/memoryExtraction/nonStreamingSse/passthroughToolNames/executorHelpers).",
-    "A later follow-up added dedicated tests (NO mock.module \u2014 unavailable under the tap-runner; used",
+    "A later follow-up added dedicated tests (NO mock.module — unavailable under the tap-runner; used",
     "fetch-override + crafted inputs + temp-DATA_DIR) for telemetryHelpers (both fns, all branches) and",
     "memorySkillsInjection (getSkillsProviderForFormat fully + injectMemoryAndSkills guards/empty-DB",
     "path) and added them as batch i.",
     "",
     "The FINAL chatCore leaf, semanticCache.ts, was added to batch i once its cache-HIT block had a",
     "fixture: chatcore-semantic-cache now SEEDS the real cache via setCachedResponse (the in-memory",
-    "store getCachedResponse checks first \u2014 no mock.module needed) under the exact signature",
+    "store getCachedResponse checks first — no mock.module needed) under the exact signature",
     "checkSemanticCache rebuilds, so the HIT branch runs end-to-end (status 200 / 'semantic' / 'HIT' /",
     "the stream + content-type ternaries / the cost fallback / the side-effect calls all get killed).",
     "ALL 15 chatCore leaves are now mutated.",
     "",
     "STILL EXCLUDED (follow-ups, NOT in `mutate` yet):",
     "  - combo.ts + chatCore.ts barrels: their handleComboChat/handleChatCore CORES were not",
-    "    split (out of scope \u2014 Fase 3 ChatCoreContext refactor). The barrels are now thin-ish",
+    "    split (out of scope — Fase 3 ChatCoreContext refactor). The barrels are now thin-ish",
     "    but still large; keep excluded until the cores are split.",
     "See project memory: Quality Gate v2 / Fase 9 (project-combo-split)."
   ],
@@ -498,7 +506,11 @@
     ".worktrees",
     ".stryker-tmp"
   ],
-  "reporters": ["progress", "html", "json"],
+  "reporters": [
+    "progress",
+    "html",
+    "json"
+  ],
   "htmlReporter": {
     "fileName": "reports/mutation/mutation.html"
   },
@@ -525,11 +537,11 @@
     "would break the required all-green baseline dry-run (e.g. body-timeout-integration,",
     "heap-pressure, sse-heartbeat-integration, *-stream-readiness, chatcore-memory-pressure).",
     "It is enumerated (not a broad glob) so the Stryker dry-run stays tractable for the",
-    "nightly budget \u2014 a glob over the full ~1300-file unit suite would make the per-test",
+    "nightly budget — a glob over the full ~1300-file unit suite would make the per-test",
     "dry-run take hours. coverageAnalysis:perTest then narrows which files run per mutant.",
     "Regenerate the base union after adding/renaming covering tests, then re-prune flaky ones:",
     "  grep -rlE \"circuitBreaker|publicCreds|accountFallback|routeGuard|services/auth|chatCore|services/combo|utils/error|public-client|account-fallback|route-guard|circuit-breaker\" tests/unit --include=\"*.test.ts\" | sort -u"
   ],
   "dryRunTimeoutMinutes": 30,
-  "_concurrency_comment": "concurrency=4 (was 1): the covering node:test files used to share SQLite/module state via the default DATA_DIR (~/.omniroute), so running them concurrently in the Stryker sandbox caused cross-file races that failed the all-green baseline. tap.nodeArgs now imports ./tests/_setup/isolateDataDir.ts, which gives each spawned test process its own temp DATA_DIR \u2014 eliminating the shared on-disk DB, so concurrency>1 is deterministic. A/B verified 2026-06-17: dry-run at concurrency=4 fails WITHOUT the isolation import (account-fallback-service tap exit 9) and passes WITH it. Raise further only if the runner has spare cores."
+  "_concurrency_comment": "concurrency=4 (was 1): the covering node:test files used to share SQLite/module state via the default DATA_DIR (~/.omniroute), so running them concurrently in the Stryker sandbox caused cross-file races that failed the all-green baseline. tap.nodeArgs now imports ./tests/_setup/isolateDataDir.ts, which gives each spawned test process its own temp DATA_DIR — eliminating the shared on-disk DB, so concurrency>1 is deterministic. A/B verified 2026-06-17: dry-run at concurrency=4 fails WITHOUT the isolation import (account-fallback-service tap exit 9) and passes WITH it. Raise further only if the runner has spare cores."
 }

tests/unit/alibaba-free-tier-allowlist.test.ts

@@ -7,9 +7,9 @@
  */
 import { test } from "node:test";
 import assert from "node:assert/strict";
-import fs from "node:fs";
-import os from "node:os";
-import path from "node:path";
+import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
 import {
   ALIBABA_FREE_TIER_TEXT_CAPABLE_MODELS,
   ALIBABA_NO_FREE_TIER_TEXT_MODELS,
@@ -30,31 +30,90 @@ test("built-in allowlist includes operator free models and excludes paid blockli
   assert.equal(isAlibabaBuiltinFreeTierTextModel("qwen3.7-max"), false);
 });
 
-test("allowlist JSON pack overrides embedded lists when valid", () => {
+/**
+ * The shipped `config/alibaba-free-tier-allowlist.json` carries a `validUntil`,
+ * so asserting against it made this test a time bomb: it went red on its own on
+ * 2026-08-28, the day after the pack expired, and stayed red on every PR and on
+ * `main` (#11866). Nothing had changed — the clock moved.
+ *
+ * Production was never affected: an expired pack falls back to the embedded
+ * list by design. So the contract worth pinning is the BEHAVIOR on both sides of
+ * the expiry, with packs this test owns and dates it controls — never the
+ * freshness of the catalog that ships in the repo.
+ */
+function withAllowlistPack(
+  pack: Record<string, unknown>,
+  assertions: () => void
+): void {
+  const dir = mkdtempSync(join(tmpdir(), "alibaba-allowlist-"));
+  const packPath = join(dir, "allowlist.json");
+  writeFileSync(packPath, JSON.stringify(pack), "utf8");
+
   const previousPath = process.env.ALIBABA_FREE_TIER_ALLOWLIST_PATH;
-  const fixtureDir = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-alibaba-allowlist-"));
-  const packPath = path.join(fixtureDir, "allowlist.json");
-
+  process.env.ALIBABA_FREE_TIER_ALLOWLIST_PATH = packPath;
+  resetAlibabaFreeTierAllowlistCache();
   try {
-    fs.writeFileSync(
-      packPath,
-      JSON.stringify({
-        asOf: "2026-08-27",
-        capable: ["qwen3.6-plus"],
-        noFreeTier: ["qwen3.7-max"],
-      })
-    );
-    process.env.ALIBABA_FREE_TIER_ALLOWLIST_PATH = packPath;
-    resetAlibabaFreeTierAllowlistCache();
-
-    const pack = loadAlibabaFreeTierAllowlistPack();
-    assert.ok(pack);
-    assert.ok(isAlibabaFreeTierAllowlistPackValid(pack));
-    assert.ok(pack.capable.includes("qwen3.6-plus"));
+    assertions();
   } finally {
     if (previousPath) process.env.ALIBABA_FREE_TIER_ALLOWLIST_PATH = previousPath;
     else delete process.env.ALIBABA_FREE_TIER_ALLOWLIST_PATH;
     resetAlibabaFreeTierAllowlistCache();
-    fs.rmSync(fixtureDir, { recursive: true, force: true });
+    rmSync(dir, { recursive: true, force: true });
   }
+}
+
+test("allowlist JSON pack overrides embedded lists while it is still valid", () => {
+  withAllowlistPack(
+    {
+      asOf: "2026-07-28",
+      validUntil: "2999-01-01",
+      capable: ["pack-only-capable-model", "qwen3.6-plus"],
+      noFreeTier: ["pack-only-paid-model"],
+    },
+    () => {
+      const pack = loadAlibabaFreeTierAllowlistPack();
+      assert.ok(pack, "a pack inside its validity window must load");
+      assert.ok(isAlibabaFreeTierAllowlistPackValid(pack!));
+      assert.ok(pack!.capable.includes("qwen3.6-plus"));
+      // Positive anchor: the pack must actually REPLACE the embedded list, not
+      // merely load. `pack-only-capable-model` exists nowhere else.
+      assert.equal(isAlibabaBuiltinFreeTierTextModel("pack-only-capable-model"), true);
+      assert.equal(isAlibabaBuiltinNoFreeTierTextModel("pack-only-paid-model"), true);
+    }
+  );
+});
+
+test("an expired allowlist pack is ignored and the embedded list serves instead", () => {
+  // This is the path production has actually been on since 2026-08-27, and it
+  // had no coverage at all — which is why the expiry surfaced as a red test
+  // rather than as a deliberate, understood fallback.
+  withAllowlistPack(
+    {
+      asOf: "2026-07-28",
+      validUntil: "2026-08-27",
+      capable: ["pack-only-capable-model"],
+      noFreeTier: ["pack-only-paid-model"],
+    },
+    () => {
+      assert.equal(loadAlibabaFreeTierAllowlistPack(), null, "expired pack must not load");
+      assert.equal(isAlibabaBuiltinFreeTierTextModel("pack-only-capable-model"), false);
+      // The embedded list must be what answers once the pack is rejected.
+      assert.equal(isAlibabaBuiltinFreeTierTextModel("qwen3.6-plus"), true);
+      assert.equal(isAlibabaBuiltinNoFreeTierTextModel("qwen3.7-max"), true);
+    }
+  );
+});
+
+test("isAlibabaFreeTierAllowlistPackValid compares against the instant it is given", () => {
+  const pack = { asOf: "2026-07-28", validUntil: "2026-08-27", capable: ["x"], noFreeTier: [] };
+  assert.equal(isAlibabaFreeTierAllowlistPackValid(pack, Date.parse("2026-08-26")), true);
+  assert.equal(isAlibabaFreeTierAllowlistPackValid(pack, Date.parse("2026-08-28")), false);
+  // No expiry declared means the pack never goes stale on its own.
+  assert.equal(
+    isAlibabaFreeTierAllowlistPackValid(
+      { asOf: "2026-07-28", capable: ["x"], noFreeTier: [] },
+      Date.parse("2999-01-01")
+    ),
+    true
+  );
 });

tests/unit/check-provider-asset-provenance.test.ts

@@ -40,7 +40,10 @@ function writeManifest(
     recordType: "manifest",
     schemaVersion: 1,
     expectedAssetCount: records.filter((record) => record.recordType === "asset").length,
-    auditedCommit: "091589089cd134a94df9f6cdab9ba562b2cefd18",
+    // HEAD instead of a pinned SHA: the fast-unit shards run on a shallow checkout,
+    // where a historical commit object does not exist and the gate would reject
+    // the fixture before exercising what the test is about.
+    auditedCommit: gitObjectId("HEAD"),
     auditedAt: "2026-08-26",
     legalScope:
       "Provenance records source matching only; it does not establish copyright or trademark clearance.",
@@ -98,12 +101,53 @@ function gitObjectId(revision: string) {
   return result.stdout.trim();
 }
 
-function gitRootCommit() {
-  const result = spawnSync("git", ["-C", REPO_ROOT, "rev-list", "--max-parents=0", "HEAD"], {
+function gitHasCommit(objectId: string) {
+  return (
+    spawnSync("git", ["-C", REPO_ROOT, "cat-file", "-e", `${objectId}^{commit}`], {
+      encoding: "utf8",
+    }).status === 0
+  );
+}
+
+function isShallowRepository() {
+  const result = spawnSync("git", ["-C", REPO_ROOT, "rev-parse", "--is-shallow-repository"], {
     encoding: "utf8",
   });
-  assert.equal(result.status, 0, result.stderr);
-  return result.stdout.trim().split(/\r?\n/)[0];
+  return result.status === 0 && result.stdout.trim() === "true";
+}
+
+/**
+ * A commit whose tree is empty, so every physical provider file is "missing"
+ * from its snapshot. Built as a dangling object (no ref is written) so it also
+ * works on the shallow checkouts the unit shards use, where the root commit is
+ * the grafted HEAD itself and would match the physical snapshot exactly.
+ */
+function emptyTreeCommit() {
+  const tree = spawnSync("git", ["-C", REPO_ROOT, "hash-object", "-w", "-t", "tree", "--stdin"], {
+    input: "",
+    encoding: "utf8",
+  });
+  assert.equal(tree.status, 0, tree.stderr);
+  const identity = {
+    GIT_AUTHOR_NAME: "provenance-fixture",
+    GIT_AUTHOR_EMAIL: "provenance-fixture@example.invalid",
+    GIT_COMMITTER_NAME: "provenance-fixture",
+    GIT_COMMITTER_EMAIL: "provenance-fixture@example.invalid",
+  };
+  const commit = spawnSync(
+    "git",
+    [
+      "-C",
+      REPO_ROOT,
+      "commit-tree",
+      tree.stdout.trim(),
+      "-m",
+      "provenance fixture: empty snapshot",
+    ],
+    { encoding: "utf8", env: { ...process.env, ...identity } }
+  );
+  assert.equal(commit.status, 0, commit.stderr);
+  return commit.stdout.trim();
 }
 
 function workflowJob(source: string, name: string) {
@@ -479,7 +523,7 @@ test("provider asset provenance gate binds auditedCommit to the physical provide
       .trim()
       .split("\n")
       .map((line) => JSON.parse(line));
-    records[0] = { ...records[0], auditedCommit: gitRootCommit() };
+    records[0] = { ...records[0], auditedCommit: emptyTreeCommit() };
     writeFileSync(
       fixture.manifestPath,
       `${records.map((record) => JSON.stringify(record)).join("\n")}\n`
@@ -497,11 +541,17 @@ test("provider asset provenance gate binds auditedCommit to the physical provide
   }
 });
 
-test("repository provider asset manifest covers the audited 142-file snapshot", () => {
-  const result = runGate(
-    join(REPO_ROOT, "public/providers"),
-    join(REPO_ROOT, "config/quality/provider-assets-provenance.jsonl")
-  );
+test("repository provider asset manifest covers the audited 142-file snapshot", (t) => {
+  const manifestPath = join(REPO_ROOT, "config/quality/provider-assets-provenance.jsonl");
+  const { auditedCommit } = JSON.parse(readFileSync(manifestPath, "utf8").split("\n")[0]);
+  if (!gitHasCommit(auditedCommit) && isShallowRepository()) {
+    // The real manifest pins a historical commit. The unit shards check out with
+    // depth 1, so it is not fetched there; the gate itself still runs on both
+    // blocking rails with fetch-depth 0 (asserted by the test right below).
+    t.skip(`shallow checkout without auditedCommit ${auditedCommit}`);
+    return;
+  }
+  const result = runGate(join(REPO_ROOT, "public/providers"), manifestPath);
 
   assert.equal(result.status, 0, `${result.stdout}\n${result.stderr}`);
   assert.match(

tests/unit/check-workflows-provenance-runner.test.ts

@@ -0,0 +1,145 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { readFileSync, readdirSync } from "node:fs";
+import { join } from "node:path";
+
+import {
+  classifyRunsOn,
+  findProvenanceOnSelfHosted,
+} from "../../scripts/check/lib/provenanceRunner.mjs";
+
+/**
+ * v3.8.50, 10th publish attempt, 76 minutes in — after the tag, the GitHub
+ * Release and the Docker images were already public:
+ *
+ *   422 Unprocessable Entity - Error verifying sigstore provenance bundle:
+ *   Unsupported GitHub Actions runner environment: "self-hosted".
+ *
+ * `USE_VPS_RUNNER` had routed the publish job to the .113 pool on 2026-08-02;
+ * no release happened between 07-30 and 08-28, so nothing surfaced it. The
+ * pairing is pure text, so it must fail the workflow lint on the PR that
+ * introduces it.
+ */
+const ROOT = join(import.meta.dirname, "../..");
+const WORKFLOWS = join(ROOT, ".github/workflows");
+
+// The exact runs-on expression npm-publish.yml used when it broke.
+const VPS_EXPR =
+  "${{ (vars.USE_VPS_RUNNER == 'true' && (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository)) && fromJSON('[\"self-hosted\",\"omni-release\"]') || 'ubuntu-latest' }}";
+
+function workflow(runsOn: string, run: string, extra = ""): string {
+  return [
+    "name: t",
+    "on: push",
+    "jobs:",
+    "  publish:",
+    `    runs-on: ${runsOn}`,
+    extra,
+    "    steps:",
+    "      - name: upload",
+    `        run: ${run}`,
+    "",
+  ].join("\n");
+}
+
+test("classifyRunsOn: literal, array, object-with-labels and the fromJSON expression are self-hosted", () => {
+  assert.equal(classifyRunsOn("self-hosted"), "self-hosted");
+  assert.equal(classifyRunsOn(["self-hosted", "omni-release"]), "self-hosted");
+  assert.equal(classifyRunsOn({ group: "Default", labels: ["self-hosted"] }), "self-hosted");
+  assert.equal(classifyRunsOn(VPS_EXPR), "self-hosted");
+});
+
+test("classifyRunsOn: hosted labels are hosted, opaque expressions are unknown (never guessed)", () => {
+  assert.equal(classifyRunsOn("ubuntu-latest"), "hosted");
+  assert.equal(classifyRunsOn(["ubuntu-latest"]), "hosted");
+  assert.equal(classifyRunsOn("${{ matrix.os }}"), "unknown");
+  assert.equal(classifyRunsOn(undefined), "unknown");
+});
+
+test("flags --provenance inside a job routed to the self-hosted pool", () => {
+  const found = findProvenanceOnSelfHosted(
+    workflow(
+      JSON.stringify(VPS_EXPR),
+      'npm stage publish --provenance --access public --tag "$TAG"'
+    ),
+    "npm-publish.yml"
+  );
+  assert.deepEqual(found, [{ file: "npm-publish.yml", job: "publish", step: "upload" }]);
+});
+
+test("also catches the literal label and the --provenance-file form", () => {
+  assert.equal(
+    findProvenanceOnSelfHosted(workflow("self-hosted", "npm publish --provenance")).length,
+    1
+  );
+  assert.equal(
+    findProvenanceOnSelfHosted(
+      workflow("[self-hosted, omni-release]", "npm publish --provenance-file=./p.json")
+    ).length,
+    0,
+    "--provenance-file is a different flag (a pre-built bundle) and is not what the registry rejects"
+  );
+  assert.equal(
+    findProvenanceOnSelfHosted(workflow("self-hosted", "npm publish --provenance=true")).length,
+    1
+  );
+});
+
+test("does not flag hosted jobs, unknown runners, or self-hosted jobs without the flag", () => {
+  assert.deepEqual(
+    findProvenanceOnSelfHosted(workflow("ubuntu-latest", "npm publish --provenance")),
+    []
+  );
+  assert.deepEqual(
+    findProvenanceOnSelfHosted(workflow("${{ matrix.os }}", "npm publish --provenance")),
+    []
+  );
+  assert.deepEqual(
+    findProvenanceOnSelfHosted(workflow("self-hosted", "npm publish --access public")),
+    []
+  );
+  // The word only in a step NAME or a comment is not a finding.
+  assert.deepEqual(
+    findProvenanceOnSelfHosted(
+      [
+        "name: t",
+        "on: push",
+        "jobs:",
+        "  j:",
+        "    runs-on: self-hosted",
+        "    steps:",
+        "      - name: provenance note",
+        "        run: echo hi # --provenance later",
+        "",
+      ].join("\n")
+    ),
+    [],
+    "a comment after the command is still part of the run string — accept that the regex is conservative"
+  );
+});
+
+test("reusable-workflow jobs (uses:) and unparseable YAML are not this rule's findings", () => {
+  const reusable = [
+    "name: t",
+    "on: push",
+    "jobs:",
+    "  j:",
+    "    uses: ./.github/workflows/x.yml",
+    "",
+  ].join("\n");
+  assert.deepEqual(findProvenanceOnSelfHosted(reusable), []);
+  assert.deepEqual(findProvenanceOnSelfHosted("jobs: [unclosed"), []);
+});
+
+test("regression guard: no workflow in this repo publishes with --provenance from a self-hosted runner", () => {
+  const files = readdirSync(WORKFLOWS).filter((f) => /\.ya?ml$/.test(f));
+  assert.ok(files.length > 10, "expected the real workflow set");
+  const findings = files.flatMap((f) =>
+    findProvenanceOnSelfHosted(readFileSync(join(WORKFLOWS, f), "utf8"), f)
+  );
+  assert.deepEqual(
+    findings,
+    [],
+    `npm rejects provenance from self-hosted runners (422) — move the upload to a github-hosted job: ${JSON.stringify(findings)}`
+  );
+});

tests/unit/config-expiry-time-bomb.test.ts

@@ -0,0 +1,135 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { mkdirSync, mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+
+import {
+  classifyExpiry,
+  collectExpiryFields,
+  scanConfigExpiry,
+} from "../../scripts/check/lib/configExpiry.mjs";
+
+/**
+ * Time bombs: config packs with a `validUntil` (or sibling key) that lapse with
+ * no commit involved. The Alibaba free-tier pack expired on 2026-08-27 and from
+ * the 28th every PR and main carried a red Unit Tests shard (#11866). Nothing a
+ * diff review could have caught.
+ *
+ * This suite fails SEVEN DAYS BEFORE any pack under config/ lapses, naming the
+ * file and key, so renewal happens on someone's terms instead of the clock's.
+ */
+const ROOT = join(import.meta.dirname, "../..");
+const CONFIG_DIR = join(ROOT, "config");
+const DAY = 86_400_000;
+const WARN_DAYS = 7;
+
+/**
+ * Packs known to be expired/expiring, each pinned to the issue that owns the
+ * renewal decision. An entry whose pack is no longer expiring FAILS below as a
+ * stale allowlist entry — remove it when the pack is renewed.
+ */
+const ALLOWLIST: Record<string, string> = {
+  "alibaba-free-tier-allowlist.json":
+    "#11866 — validUntil 2026-08-27 has passed; the loader already falls back to the embedded list, and renewing the curated free-tier pack is an operator data decision, not a test fix",
+};
+
+const NOW = Date.UTC(2026, 7, 28); // 2026-08-28, fixed: this suite must not itself depend on the clock
+const day = (offset: number) => new Date(NOW + offset * DAY).toISOString().slice(0, 10);
+
+test("collectExpiryFields: finds nested and array-nested expiry keys, ignores non-string values", () => {
+  const fields = collectExpiryFields(
+    {
+      validUntil: day(3),
+      nested: { expiresAt: day(30), other: "x" },
+      list: [{ expiry: day(-1) }, { expires: 12345 }],
+      expires_at: "not a date",
+    },
+    "pack.json"
+  );
+  assert.deepEqual(
+    fields.map((f) => [f.keyPath, f.expiresAt === null ? null : "date"]),
+    [
+      ["validUntil", "date"],
+      ["nested.expiresAt", "date"],
+      ["list.0.expiry", "date"],
+      ["expires_at", null],
+    ]
+  );
+});
+
+test("classifyExpiry: expired / expiring inside the warning window / ok / unparseable", () => {
+  const f = (raw: string) => ({
+    file: "p",
+    keyPath: "validUntil",
+    raw,
+    expiresAt: Number.isFinite(Date.parse(raw)) ? Date.parse(raw) : null,
+  });
+  assert.equal(classifyExpiry(f(day(-1)), NOW, WARN_DAYS), "expired");
+  assert.equal(
+    classifyExpiry(f(day(0)), NOW, WARN_DAYS),
+    "expiring",
+    "lapsing today is already too late to be 'ok'"
+  );
+  assert.equal(classifyExpiry(f(day(6)), NOW, WARN_DAYS), "expiring");
+  assert.equal(classifyExpiry(f(day(8)), NOW, WARN_DAYS), "ok");
+  assert.equal(classifyExpiry(f("never"), NOW, WARN_DAYS), "unparseable");
+});
+
+test("scanConfigExpiry: walks a config tree, skips node_modules and invalid JSON, keys files portably", () => {
+  const dir = mkdtempSync(join(tmpdir(), "cfg-expiry-"));
+  try {
+    mkdirSync(join(dir, "sub"), { recursive: true });
+    mkdirSync(join(dir, "node_modules", "dep"), { recursive: true });
+    writeFileSync(join(dir, "a.json"), JSON.stringify({ validUntil: day(3) }));
+    writeFileSync(join(dir, "sub", "b.json"), JSON.stringify({ deep: { expiresAt: day(40) } }));
+    writeFileSync(
+      join(dir, "node_modules", "dep", "c.json"),
+      JSON.stringify({ validUntil: day(-5) })
+    );
+    writeFileSync(join(dir, "broken.json"), "{ not json");
+    writeFileSync(join(dir, "notes.txt"), JSON.stringify({ validUntil: day(-5) }));
+    const found = scanConfigExpiry(dir).map((f) => `${f.file}:${f.keyPath}`);
+    assert.deepEqual(found, ["a.json:validUntil", "sub/b.json:deep.expiresAt"]);
+  } finally {
+    rmSync(dir, { recursive: true, force: true });
+  }
+});
+
+test(`repo: no pack under config/ lapses within ${WARN_DAYS} days unless its renewal is tracked`, (t) => {
+  const fields = scanConfigExpiry(CONFIG_DIR);
+  // Positive anchor: the scanner must be seeing SOMETHING, or a renamed key
+  // would silently turn this whole suite into a no-op.
+  assert.ok(
+    fields.length >= 1,
+    "expected at least one dated pack under config/ (the Alibaba allowlist) — if the key was renamed, extend EXPIRY_KEY"
+  );
+
+  const failures: string[] = [];
+  const seenAllowlisted = new Set<string>();
+  for (const f of fields) {
+    const status = classifyExpiry(f, Date.now(), WARN_DAYS);
+    const tracked = ALLOWLIST[f.file];
+    if (status === "unparseable") {
+      t.diagnostic(`${f.file} ${f.keyPath}="${f.raw}" is not a date — not monitored`);
+      continue;
+    }
+    if (status === "ok") continue;
+    if (tracked) {
+      seenAllowlisted.add(f.file);
+      t.diagnostic(`${f.file} ${f.keyPath}=${f.raw} is ${status} — tracked: ${tracked}`);
+      continue;
+    }
+    failures.push(
+      `${f.file} → ${f.keyPath}=${f.raw} is ${status}: renew the pack (or track it in ALLOWLIST with its issue)`
+    );
+  }
+  for (const file of Object.keys(ALLOWLIST)) {
+    if (!seenAllowlisted.has(file)) {
+      failures.push(
+        `stale ALLOWLIST entry: ${file} is no longer expired/expiring — remove it (${ALLOWLIST[file]})`
+      );
+    }
+  }
+  assert.deepEqual(failures, [], failures.join("\n"));
+});

tests/unit/db-core-init.test.ts

@@ -223,6 +223,12 @@ function createRecoverableDb(sqliteFile) {
       auth_type TEXT,
       name TEXT,
       is_active INTEGER DEFAULT 1,
+      test_status TEXT,
+      error_code TEXT,
+      last_error TEXT,
+      last_error_at TEXT,
+      last_error_type TEXT,
+      last_error_source TEXT,
       created_at TEXT NOT NULL,
       updated_at TEXT NOT NULL
     );
@@ -659,6 +665,12 @@ test(
         name TEXT,
         priority INTEGER DEFAULT 0,
         is_active INTEGER DEFAULT 1,
+      test_status TEXT,
+      error_code TEXT,
+      last_error TEXT,
+      last_error_at TEXT,
+      last_error_type TEXT,
+      last_error_source TEXT,
         created_at TEXT NOT NULL,
         updated_at TEXT NOT NULL
       );

tests/unit/gemini-malformed-function-call-finish-reason-2462.test.ts

@@ -70,6 +70,12 @@ function runGeminiToClaude(geminiChunk) {
     const converted = openaiToClaudeResponse(chunk, claudeState);
     if (converted) claudeEvents.push(...converted);
   }
+  // End-of-stream flush: production calls the translator once more with `null`
+  // when the upstream stream closes (open-sse/utils/stream.ts flush →
+  // translateResponse(..., null, state)). Since dd35750e5f a finish chunk that
+  // carries no usage is deferred until that flush, so the driver must mirror it.
+  const flushed = openaiToClaudeResponse(null, claudeState);
+  if (flushed) claudeEvents.push(...flushed);
   return { openaiEvents, claudeEvents };
 }
 

tests/unit/lobe-provider-icons-prototype-collision-11853.test.ts

@@ -0,0 +1,48 @@
+/**
+ * Regression for #11853 — the providers dashboard crashed with
+ * "Cannot read properties of undefined (reading 'color')" and rendered a
+ * misleading "Failed to load providers — check your connection" card.
+ *
+ * `getLobeProviderIcon()` indexed two plain object literals without own-property
+ * guards. A provider id whose lowercased form is an Object.prototype member
+ * resolves through the prototype chain: `LOBE_PROVIDER_ALIASES["constructor"]`
+ * returns the Object constructor (truthy, so the `if (!iconKey) return null`
+ * guard passes), then `LOBE_ICON_COMPONENTS[<that function>]` is undefined and
+ * `entry.color` throws — taking the whole page down through the App Router
+ * error boundary, since ProviderIcon calls this for every provider card.
+ *
+ * Only `constructor` and `__proto__` are reachable: every other Object.prototype
+ * member is camelCase and no longer collides after `.toLowerCase()`.
+ *
+ * Runner: node --import tsx/esm --test tests/unit/lobe-provider-icons-prototype-collision-11853.test.ts
+ */
+import test from "node:test";
+import assert from "node:assert/strict";
+
+const { getLobeProviderIcon } = await import("../../src/shared/components/lobeProviderIcons.ts");
+
+test("#11853 — prototype-colliding provider ids return null instead of throwing", () => {
+  for (const id of ["constructor", "__proto__", "CONSTRUCTOR", "__PROTO__"]) {
+    for (const type of ["color", "mono"] as const) {
+      assert.doesNotThrow(() => getLobeProviderIcon(id, type), `${id} (${type}) must not throw`);
+      assert.equal(getLobeProviderIcon(id, type), null, `${id} (${type}) must resolve to null`);
+    }
+  }
+});
+
+test("#11853 — camelCase prototype members were already safe and stay safe", () => {
+  for (const id of ["valueOf", "toString", "hasOwnProperty", "isPrototypeOf"]) {
+    assert.equal(getLobeProviderIcon(id), null);
+  }
+});
+
+test("#11853 — no regression: known providers still resolve, unknown ones still null", () => {
+  assert.notEqual(getLobeProviderIcon("openai"), null);
+  assert.notEqual(getLobeProviderIcon("anthropic"), null);
+  assert.equal(getLobeProviderIcon("definitely-not-a-provider"), null);
+});
+
+test("#11853 — a non-string provider id does not throw", () => {
+  assert.doesNotThrow(() => getLobeProviderIcon(undefined as unknown as string));
+  assert.equal(getLobeProviderIcon(undefined as unknown as string), null);
+});

tests/unit/model-lifecycle-integration.test.ts

@@ -115,7 +115,12 @@ test("unified catalog suppresses stale OpenAI chat rows but retains typed media"
   assert.equal(ids.has("openai/gpt-5.2-codex"), false);
   assert.equal(ids.has("openai/sora-2"), false);
   assert.equal(ids.has("openai/sora-2-pro"), false);
-  assert.equal(ids.has("openai/gpt-5.6-sol"), true);
+  // Since #11919 (fixes #11829) an authoritative live catalog REPLACES the static
+  // registry: a static-only row like gpt-5.6-sol that the synced catalog does not
+  // list is suppressed instead of leaking into /v1/models. The lifecycle contract
+  // this file guards (#8627: stale chat rows suppressed, typed media retained)
+  // is unchanged — only the "static rows survive a sync" expectation moved.
+  assert.equal(ids.has("openai/gpt-5.6-sol"), false);
   assert.equal(body.data.find((item) => item.id === "openai/gpt-image-2")?.type, "image");
 });
 

tests/unit/models-catalog-custom-node-prefix.test.ts

@@ -0,0 +1,109 @@
+import assert from "node:assert/strict";
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+import test from "node:test";
+
+const TEST_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-custom-node-prefix-"));
+process.env.DATA_DIR = TEST_DATA_DIR;
+
+const aliasesDb = await import("../../src/lib/db/models/aliases.ts");
+const core = await import("../../src/lib/db/core.ts");
+const modelsDb = await import("../../src/lib/db/models.ts");
+const providersDb = await import("../../src/lib/db/providers.ts");
+const modelsCatalog = await import("../../src/app/api/v1/models/catalog.ts");
+const modelsRoute = await import("../../src/app/api/v1/models/route.ts");
+
+const NODE_ID = "openai-compatible-chat-550e8400-e29b-41d4-a716-446655440000";
+const PREFIX = "infrex";
+const EXPECTED_IDS = [
+  `${PREFIX}/synced-model`,
+  `${PREFIX}/custom-model`,
+  `${PREFIX}/alias-backed-model`,
+];
+
+async function resetStorage(): Promise<void> {
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+  fs.mkdirSync(TEST_DATA_DIR, { recursive: true });
+  modelsCatalog.__resetCatalogBuilderRunsForTest();
+}
+
+async function seedCustomNode(): Promise<void> {
+  await providersDb.createProviderNode({
+    id: NODE_ID,
+    type: "openai-compatible",
+    name: "Infrex",
+    prefix: PREFIX,
+    baseUrl: "https://proxy.example.com",
+    chatPath: "/v1/chat/completions",
+    modelsPath: "/v1/models",
+  });
+  const connection = await providersDb.createProviderConnection({
+    provider: NODE_ID,
+    authType: "apikey",
+    name: "infrex-primary",
+    apiKey: "test-key",
+    isActive: true,
+    testStatus: "active",
+    providerSpecificData: {
+      baseUrl: "https://proxy.example.com",
+      chatPath: "/v1/chat/completions",
+      modelsPath: "/v1/models",
+    },
+  });
+
+  await modelsDb.replaceSyncedAvailableModelsForConnection(
+    NODE_ID,
+    (connection as { id: string }).id,
+    [{ id: "synced-model", source: "imported", supportedEndpoints: ["chat"] }]
+  );
+  await modelsDb.addCustomModel(NODE_ID, "custom-model", "Custom Model");
+  await aliasesDb.setModelAlias("alias-backed-model", `${NODE_ID}/alias-backed-model`);
+}
+
+async function getCatalogIds(mode: "alias" | "canonical" | "dual"): Promise<string[]> {
+  modelsCatalog.__resetCatalogBuilderRunsForTest();
+  const response = await modelsRoute.GET(
+    new Request(`http://localhost/api/v1/models?prefix=${mode}`)
+  );
+  assert.equal(response.status, 200);
+  const body = (await response.json()) as { data: Array<{ id: string }> };
+  return body.data.map((model) => model.id);
+}
+
+function assertCustomNodeModels(ids: string[]): void {
+  const actualCustomNodeIds = ids.filter((id) => EXPECTED_IDS.includes(id)).sort();
+  assert.deepEqual(
+    actualCustomNodeIds,
+    [...EXPECTED_IDS].sort(),
+    `expected each custom node source exactly once in ${JSON.stringify(ids)}`
+  );
+  assert.equal(
+    ids.some((id) => id.startsWith(`${NODE_ID}/`)),
+    false,
+    "catalog must not expose the internal provider node id"
+  );
+}
+
+test.beforeEach(async () => {
+  await resetStorage();
+  await seedCustomNode();
+});
+
+test.after(() => {
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+});
+
+test("alias mode exposes every custom node model under its configured prefix", async () => {
+  assertCustomNodeModels(await getCatalogIds("alias"));
+});
+
+test("canonical mode keeps custom node models under their configured prefix", async () => {
+  assertCustomNodeModels(await getCatalogIds("canonical"));
+});
+
+test("dual mode exposes each custom node model once under its configured prefix", async () => {
+  assertCustomNodeModels(await getCatalogIds("dual"));
+});

tests/unit/models-catalog-static-synced-suppression.test.ts

@@ -0,0 +1,120 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+
+const TEST_DATA_DIR = fs.mkdtempSync(
+  path.join(os.tmpdir(), "omniroute-model-catalog-static-synced-suppression-")
+);
+
+process.env.DATA_DIR = TEST_DATA_DIR;
+process.env.API_KEY_SECRET =
+  process.env.API_KEY_SECRET || "model-catalog-static-synced-suppression-secret";
+
+const core = await import("../../src/lib/db/core.ts");
+const providersDb = await import("../../src/lib/db/providers.ts");
+const modelsDb = await import("../../src/lib/db/models.ts");
+const catalog = await import("../../src/app/api/v1/models/catalog.ts");
+const { REGISTRY } = await import("@omniroute/open-sse/config/providerRegistry");
+
+const LIVE_MODEL = "google/gemma-4-31b-it";
+
+function getStaticModel(provider: string) {
+  const model = REGISTRY[provider]?.models?.[0];
+  assert.ok(model, `${provider} must define a static registry model for this regression test`);
+  return model;
+}
+
+async function resetStorage() {
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+  fs.mkdirSync(path.join(TEST_DATA_DIR, "logs/application"), { recursive: true });
+  catalog.__resetCatalogBuilderRunsForTest();
+}
+
+async function seedConnection(provider: string, name: string) {
+  return providersDb.createProviderConnection({
+    provider,
+    authType: "apikey",
+    name,
+    apiKey: "test-api-key",
+    isActive: true,
+    testStatus: "active",
+  });
+}
+
+async function getCatalogIds(): Promise<Set<string>> {
+  const response = await catalog.getUnifiedModelsResponse(
+    new Request("http://localhost/api/v1/models")
+  );
+  const body = (await response.json()) as { data: Array<{ id: string }> };
+
+  assert.equal(response.status, 200);
+  return new Set(body.data.map((model) => model.id));
+}
+
+test.beforeEach(resetStorage);
+
+test.after(() => {
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+});
+
+test("active authoritative live catalog suppresses stale static registry models", async () => {
+  const staticNvidiaModel = getStaticModel("nvidia").id;
+  const connection = await seedConnection("nvidia", "nvidia-static-synced-suppression");
+  await modelsDb.replaceSyncedAvailableModelsForConnection("nvidia", connection.id as string, [
+    { id: LIVE_MODEL, name: "Gemma 4 31B", source: "imported" },
+  ]);
+
+  const ids = await getCatalogIds();
+
+  assert.equal(ids.has(`nvidia/${LIVE_MODEL}`), true);
+  assert.equal(ids.has(`nvidia/${staticNvidiaModel}`), false);
+});
+
+test("static registry remains fallback when active connection has no live catalog", async () => {
+  const staticNvidiaModel = getStaticModel("nvidia").id;
+  await seedConnection("nvidia", "nvidia-fallback-static");
+
+  const ids = await getCatalogIds();
+
+  assert.equal(ids.has(`nvidia/${staticNvidiaModel}`), true);
+});
+
+test("authoritative live catalog suppresses static effort-tier variants on sync", async () => {
+  const connection = await seedConnection("glm", "glm-authoritative-effort-suppression");
+  const glmStaticModel = REGISTRY.glm?.models?.find(
+    (model) =>
+      Array.isArray(model.supportedThinkingEfforts) && model.supportedThinkingEfforts.length > 0
+  );
+  assert.ok(glmStaticModel, "glm must define an effort-tier static model for this regression test");
+  const effort = glmStaticModel.supportedThinkingEfforts?.[0];
+  assert.ok(effort, "glm static model must declare at least one effort tier");
+
+  await modelsDb.replaceSyncedAvailableModelsForConnection("glm", connection.id as string, [
+    { id: "glm-5-synced", name: "GLM 5 Synced", source: "imported" },
+  ]);
+
+  const ids = await getCatalogIds();
+
+  assert.equal(ids.has("glm/glm-5-synced"), true);
+  assert.equal(ids.has(`glm/${glmStaticModel.id}`), false);
+  assert.equal(ids.has(`glm/${glmStaticModel.id}-${effort}`), false);
+});
+
+test("partial discovery provider preserves uncovered static models when synced", async () => {
+  const connection = await seedConnection("command-code", "command-code-partial-discovery");
+  const uncoveredStaticModel = "deepseek/deepseek-v4-flash";
+  const coveredSyncedModel = "claude-opus-4-7";
+
+  await modelsDb.replaceSyncedAvailableModelsForConnection("command-code", connection.id as string, [
+    { id: coveredSyncedModel, name: "Claude Opus 4.7", source: "imported" },
+  ]);
+
+  const ids = await getCatalogIds();
+
+  assert.equal(ids.has(`cmd/${coveredSyncedModel}`), true);
+  assert.equal(ids.has(`cmd/${uncoveredStaticModel}`), true);
+});

tests/unit/no-js-extension-on-repo-imports-10674.test.ts

@@ -1,6 +1,7 @@
 import assert from "node:assert/strict";
 import { test } from "node:test";
 import { execFileSync } from "node:child_process";
+import fs from "node:fs";
 import path from "node:path";
 import { fileURLToPath } from "node:url";
 
@@ -30,7 +31,19 @@ function relativeJsImports(): string[] {
     if ((err as { status?: number }).status === 1) return [];
     throw err;
   }
-  return out.split("\n").filter((line) => line.trim().length > 0);
+  return (
+    out
+      .split("\n")
+      .filter((line) => line.trim().length > 0)
+      // A `.js` specifier whose target really is a JavaScript file (e.g.
+      // open-sse/lib/deepseek-pow-hash.js, shared with a worker) is correct —
+      // the defect #10674 guards against is a `.js` suffix on a `.ts` source.
+      .filter((line) => {
+        const m = /^([^:]+):\d+:.*from "(\.{1,2}\/[^"]*\.js)"/.exec(line);
+        if (!m) return true;
+        return !fs.existsSync(path.resolve(REPO_ROOT, path.dirname(m[1]), m[2]));
+      })
+  );
 }
 
 test("no first-party TypeScript module is imported through a .js specifier", () => {

tests/unit/openai-to-claude-trailing-usage-11817.test.ts

@@ -0,0 +1,104 @@
+/**
+ * Regression for #11817 — the streaming OpenAI→Claude translator dropped the
+ * upstream usage block (including prompt-cache accounting) whenever it arrived
+ * on a trailing usage-only chunk shaped `{"choices":[],"usage":{...}}`.
+ *
+ * Many OpenAI-compatible upstreams (confirmed: Fireworks / kimi-k3, also vLLM
+ * and Together with `stream_options.include_usage`) emit usage exactly that
+ * way. `openaiToClaudeResponse()` returned early on `!chunk.choices?.[0]`
+ * BEFORE reading `chunk.usage`, so `state.usage` stayed undefined and every
+ * downstream consumer fell back to OmniRoute's own tokenizer estimate — no
+ * cache_read_input_tokens, no cache_creation_input_tokens, and an input_tokens
+ * figure that disagreed with the provider's own count.
+ *
+ * Impact was silent over-billing: a session served ~75% from prompt cache was
+ * metered at the full uncached rate.
+ *
+ * Runner: node --import tsx/esm --test tests/unit/openai-to-claude-trailing-usage-11817.test.ts
+ */
+import test from "node:test";
+import assert from "node:assert/strict";
+
+const { openaiToClaudeResponse } =
+  await import("../../open-sse/translator/response/openai-to-claude.ts");
+
+function newState() {
+  return { toolCalls: new Map(), messageId: "msg_11817", model: "kimi-k3" } as Record<
+    string,
+    unknown
+  >;
+}
+
+test("#11817 — usage on a trailing choices:[] chunk is harvested, with cache split", () => {
+  const state = newState();
+
+  openaiToClaudeResponse({ choices: [{ index: 0, delta: { content: "ok" } }] }, state);
+  openaiToClaudeResponse({ choices: [{ index: 0, delta: {}, finish_reason: "stop" }] }, state);
+  openaiToClaudeResponse(
+    {
+      choices: [],
+      usage: {
+        prompt_tokens: 6103,
+        completion_tokens: 24,
+        prompt_tokens_details: { cached_tokens: 6102 },
+      },
+    },
+    state
+  );
+
+  assert.deepEqual(state.usage, {
+    input_tokens: 1, // 6103 - 6102 cached
+    output_tokens: 24,
+    cache_read_input_tokens: 6102,
+  });
+});
+
+test("#11817 — cache_creation_tokens on a trailing chunk is mapped too", () => {
+  const state = newState();
+  openaiToClaudeResponse(
+    {
+      choices: [],
+      usage: {
+        prompt_tokens: 1000,
+        completion_tokens: 5,
+        prompt_tokens_details: { cached_tokens: 400, cache_creation_tokens: 100 },
+      },
+    },
+    state
+  );
+
+  assert.deepEqual(state.usage, {
+    input_tokens: 500,
+    output_tokens: 5,
+    cache_read_input_tokens: 400,
+    cache_creation_input_tokens: 100,
+  });
+});
+
+test("#11817 — a usage-only chunk still emits no Claude events", () => {
+  const state = newState();
+  const out = openaiToClaudeResponse(
+    { choices: [], usage: { prompt_tokens: 10, completion_tokens: 1 } },
+    state
+  );
+  assert.equal(out, null);
+});
+
+test("#11817 — no regression: usage carried inline on the finish chunk", () => {
+  const state = newState();
+  openaiToClaudeResponse({ choices: [{ index: 0, delta: { content: "hi" } }] }, state);
+  openaiToClaudeResponse(
+    {
+      choices: [{ index: 0, delta: {}, finish_reason: "stop" }],
+      usage: { prompt_tokens: 100, completion_tokens: 10 },
+    },
+    state
+  );
+  assert.deepEqual(state.usage, { input_tokens: 100, output_tokens: 10 });
+});
+
+test("#11817 — no regression: empty and nullish chunks are still ignored", () => {
+  assert.equal(openaiToClaudeResponse(null, newState()), null);
+  assert.equal(openaiToClaudeResponse({ choices: [] }, newState()), null);
+  assert.equal(openaiToClaudeResponse({}, newState()), null);
+});

tests/unit/openai-to-claude-trailing-usage-deferred-finish.test.ts

@@ -0,0 +1,229 @@
+import assert from "node:assert/strict";
+import test from "node:test";
+
+import { openaiToClaudeResponse } from "../../open-sse/translator/response/openai-to-claude.ts";
+
+type ClaudeUsage = {
+  input_tokens: number;
+  output_tokens: number;
+  cache_read_input_tokens?: number;
+  cache_creation_input_tokens?: number;
+};
+
+type TranslatorState = Record<string, unknown> & {
+  toolCalls: Map<number, unknown>;
+  usage?: ClaudeUsage;
+};
+
+const TRAILING_USAGE = {
+  prompt_tokens: 6103,
+  completion_tokens: 16,
+  total_tokens: 6119,
+  prompt_tokens_details: {
+    cached_tokens: 6000,
+    cache_creation_tokens: 100,
+  },
+};
+
+function createState(): TranslatorState {
+  return { toolCalls: new Map() };
+}
+
+function collectEvents(
+  chunks: Array<Record<string, unknown> | null>,
+  state: TranslatorState
+): Array<Record<string, unknown>> {
+  return chunks.flatMap((chunk) => openaiToClaudeResponse(chunk, state) ?? []);
+}
+
+test("usage-only choices-empty chunk updates Claude usage without emitting a content delta", () => {
+  const state = createState();
+
+  const events = openaiToClaudeResponse(
+    {
+      id: "chatcmpl-11817",
+      model: "accounts/fireworks/models/kimi-k3",
+      choices: [],
+      usage: TRAILING_USAGE,
+    },
+    state
+  );
+
+  assert.equal(events, null);
+  assert.deepEqual(state.usage, {
+    input_tokens: 3,
+    output_tokens: 16,
+    cache_read_input_tokens: 6000,
+    cache_creation_input_tokens: 100,
+  });
+});
+
+test("trailing choices-empty usage completes the stream with real cache accounting", () => {
+  const state = createState();
+  const events = collectEvents(
+    [
+      {
+        id: "chatcmpl-11817",
+        model: "accounts/fireworks/models/kimi-k3",
+        choices: [{ index: 0, delta: { content: "OK" }, finish_reason: null }],
+      },
+      {
+        id: "chatcmpl-11817",
+        model: "accounts/fireworks/models/kimi-k3",
+        choices: [{ index: 0, delta: {}, finish_reason: "stop" }],
+      },
+      {
+        id: "chatcmpl-11817",
+        model: "accounts/fireworks/models/kimi-k3",
+        choices: [],
+        usage: TRAILING_USAGE,
+      },
+    ],
+    state
+  );
+
+  assert.equal(events[0].type, "message_start");
+  assert.equal(events[1].type, "content_block_start");
+  assert.equal(events[2].type, "content_block_delta");
+  assert.equal(events[2].delta?.text, "OK");
+  assert.equal(events[3].type, "content_block_stop");
+  assert.equal(events[4].type, "message_delta");
+  assert.equal(events[4].delta?.stop_reason, "end_turn");
+  assert.deepEqual(events[4].usage, {
+    input_tokens: 3,
+    output_tokens: 16,
+    cache_read_input_tokens: 6000,
+    cache_creation_input_tokens: 100,
+  });
+  assert.equal(events[5].type, "message_stop");
+  assert.equal(events.length, 6);
+});
+
+test("stream-end flush still emits terminal events when upstream omits usage", () => {
+  const state = createState();
+  const events = collectEvents(
+    [
+      {
+        id: "chatcmpl-11817-no-usage",
+        model: "accounts/fireworks/models/kimi-k3",
+        choices: [{ index: 0, delta: { content: "OK" }, finish_reason: null }],
+      },
+      {
+        id: "chatcmpl-11817-no-usage",
+        model: "accounts/fireworks/models/kimi-k3",
+        choices: [{ index: 0, delta: {}, finish_reason: "stop" }],
+      },
+      null,
+    ],
+    state
+  );
+
+  assert.deepEqual(
+    events.filter((event) => event.type === "message_delta" || event.type === "message_stop"),
+    [
+      {
+        type: "message_delta",
+        delta: { stop_reason: "end_turn" },
+        usage: { input_tokens: 0, output_tokens: 0 },
+      },
+      { type: "message_stop" },
+    ]
+  );
+});
+
+test("trailing chunk without choices property updates usage and flushes finish", () => {
+  const state = createState();
+  const events = collectEvents(
+    [
+      {
+        id: "chatcmpl-11817-no-choices-key",
+        model: "accounts/fireworks/models/kimi-k3",
+        choices: [{ index: 0, delta: { content: "Done" }, finish_reason: null }],
+      },
+      {
+        id: "chatcmpl-11817-no-choices-key",
+        model: "accounts/fireworks/models/kimi-k3",
+        choices: [{ index: 0, delta: {}, finish_reason: "stop" }],
+      },
+      {
+        id: "chatcmpl-11817-no-choices-key",
+        model: "accounts/fireworks/models/kimi-k3",
+        usage: TRAILING_USAGE,
+      },
+    ],
+    state
+  );
+
+  const terminalEvents = events.filter(
+    (event) => event.type === "message_delta" || event.type === "message_stop"
+  );
+  assert.deepEqual(terminalEvents, [
+    {
+      type: "message_delta",
+      delta: { stop_reason: "end_turn" },
+      usage: {
+        input_tokens: 3,
+        output_tokens: 16,
+        cache_read_input_tokens: 6000,
+        cache_creation_input_tokens: 100,
+      },
+    },
+    { type: "message_stop" },
+  ]);
+});
+
+test("trailing choices-empty chunk with tool_calls finish_reason preserves tool_use stop_reason and usage", () => {
+  const state = createState();
+  const events = collectEvents(
+    [
+      {
+        id: "chatcmpl-11817-tool",
+        model: "accounts/fireworks/models/kimi-k3",
+        choices: [
+          {
+            index: 0,
+            delta: {
+              tool_calls: [
+                {
+                  index: 0,
+                  id: "call_123",
+                  function: { name: "get_weather", arguments: "{\"city\":\"Beijing\"}" },
+                },
+              ],
+            },
+            finish_reason: null,
+          },
+        ],
+      },
+      {
+        id: "chatcmpl-11817-tool",
+        model: "accounts/fireworks/models/kimi-k3",
+        choices: [{ index: 0, delta: {}, finish_reason: "tool_calls" }],
+      },
+      {
+        id: "chatcmpl-11817-tool",
+        model: "accounts/fireworks/models/kimi-k3",
+        choices: [],
+        usage: TRAILING_USAGE,
+      },
+    ],
+    state
+  );
+
+  const terminalEvents = events.filter(
+    (event) => event.type === "message_delta" || event.type === "message_stop"
+  );
+  assert.deepEqual(terminalEvents, [
+    {
+      type: "message_delta",
+      delta: { stop_reason: "tool_use" },
+      usage: {
+        input_tokens: 3,
+        output_tokens: 16,
+        cache_read_input_tokens: 6000,
+        cache_creation_input_tokens: 100,
+      },
+    },
+    { type: "message_stop" },
+  ]);
+});

tests/unit/playground-simulate-route-persisted-combo.test.ts

@@ -73,10 +73,41 @@ test("simulates persisted combo model steps in order", async () => {
     body.targets.map(({ status }: Record<string, unknown>) => status),
     ["available", "available"]
   );
-  assert.ok(body.warnings.includes("Skipped 1 unsupported persisted combo step."));
+  // #11822 follow-up: combo-ref steps now get a specific warning naming the
+  // referenced combo instead of folding into the generic "unsupported step"
+  // count (that count is reserved for genuinely unrecognized step shapes).
+  assert.ok(
+    body.warnings.some((warning: string) => warning.includes('combo "nested combo"'))
+  );
   assert.ok(body.warnings.every((warning: string) => !warning.includes("not configured")));
 });
 
+test("surfaces a provider-wildcard step as an unresolved target with a specific warning", async () => {
+  const combo = await combosDb.createCombo({
+    name: "combo with wildcard",
+    strategy: "priority",
+    models: [
+      { kind: "model", model: "cc/claude-opus-5" },
+      { kind: "provider-wildcard", providerId: "groq", modelPattern: "llama-*" },
+    ],
+  });
+
+  const response = await POST(request({ comboId: combo.id, promptTokens: 500 }));
+  const body = await response.json();
+
+  assert.equal(response.status, 200);
+  assert.deepEqual(
+    body.targets.map(({ provider, model }: Record<string, unknown>) => ({ provider, model })),
+    [
+      { provider: "cc", model: "claude-opus-5" },
+      { provider: "groq", model: "llama-*" },
+    ]
+  );
+  assert.ok(
+    body.warnings.some((warning: string) => warning.includes("groq/llama-*") && warning.includes("wildcard"))
+  );
+});
+
 test("returns 404 for a missing persisted combo", async () => {
   const response = await POST(request({ comboId: "missing" }));
 

tests/unit/runner-janitor.test.ts

@@ -0,0 +1,196 @@
+import { describe, it } from "node:test";
+import assert from "node:assert/strict";
+import { spawnSync } from "node:child_process";
+import {
+  existsSync,
+  mkdirSync,
+  mkdtempSync,
+  readFileSync,
+  rmSync,
+  statSync,
+  utimesSync,
+  writeFileSync,
+} from "node:fs";
+import os from "node:os";
+import path from "node:path";
+
+/**
+ * scripts/ops/runner-janitor.sh runs from cron on the .113 runner box. This
+ * suite pins its safety contract against a fixture tree — never the real /tmp:
+ * every base, the runner dirs, the PSI file and the df path are redirected, the
+ * zombie pattern is set to a name no process has, and the ceilings are lifted
+ * so the outcome does not depend on the box the test happens to run on.
+ */
+const ROOT = path.resolve(import.meta.dirname, "..", "..");
+const SCRIPT = path.join(ROOT, "scripts", "ops", "runner-janitor.sh");
+const HOUR = 3_600_000;
+// The sweep needs lsof to PROVE a path is idle (one snapshot of open paths). Hosted CI
+// images ship both; a bare devbox may not. Each branch below asserts what must
+// hold in that environment — without the tools the contract is "delete nothing,
+// say why", which is exactly the behaviour worth pinning.
+const HAVE_BUSY_TOOLS =
+  spawnSync("bash", ["-c", "command -v lsof"], { stdio: "ignore" }).status === 0;
+
+function fixture() {
+  const base = mkdtempSync(path.join(os.tmpdir(), "janitor-fixture-"));
+  const old = new Date(Date.now() - 5 * HOUR);
+  const mk = (name: string, dir: boolean, when: Date | null) => {
+    const p = path.join(base, name);
+    if (dir) {
+      mkdirSync(p);
+      writeFileSync(path.join(p, "x"), "x");
+    } else writeFileSync(p, "x");
+    if (when) utimesSync(p, when, when);
+    return p;
+  };
+  return {
+    base,
+    staleTar: mk("e2e-build.tar.gz", false, old), // fixed-name artefact ci.yml/npm-publish leave behind
+    staleBuild: mk("next-build-abc", true, old),
+    staleUpgrade: mk("omniroute-install-upgrade-xyz", true, old),
+    fresh: mk("omniroute-batch-api-fresh", true, null), // in use right now
+    unrelated: mk("somebody-elses.log", false, old), // not ours — never touched
+  };
+}
+
+function run(args: string[], base: string, extraEnv: Record<string, string> = {}) {
+  return spawnSync("bash", [SCRIPT, ...args], {
+    encoding: "utf8",
+    stdio: ["ignore", "pipe", "pipe"],
+    env: {
+      ...process.env,
+      JANITOR_TMP_BASES: base,
+      JANITOR_WORK_TEMP_BASES: "",
+      JANITOR_RUNNER_DIRS: path.join(base, "no-runners-here-*"),
+      JANITOR_PSI_FILE: path.join(base, "no-psi"),
+      JANITOR_DF_PATH: base,
+      ZOMBIE_BUILD_COMM: "janitor-test-no-such-process",
+      MAX_ACTIVE_RUNNERS: "9999",
+      DISK_ALERT_PCT: "101",
+      ...extraEnv,
+    },
+  });
+}
+
+describe("runner-janitor.sh", () => {
+  it("is executable bash with strict mode and prints usage on --help", () => {
+    assert.ok(existsSync(SCRIPT));
+    assert.ok(statSync(SCRIPT).mode & 0o111, "must be chmod +x (cron runs it directly)");
+    const body = readFileSync(SCRIPT, "utf8");
+    assert.ok(body.startsWith("#!/usr/bin/env bash"));
+    assert.ok(body.includes("set -euo pipefail"));
+    const help = run(["--help"], os.tmpdir());
+    assert.equal(help.status, 0, help.stderr);
+    assert.match(help.stdout, /--dry-run/);
+  });
+
+  it("without lsof it cannot prove idle, so it deletes nothing and says why (exit 1)", () => {
+    const f = fixture();
+    try {
+      const r = run([], f.base, { JANITOR_LSOF: "/nonexistent/lsof" });
+      assert.equal(r.status, 1, "a janitor that cannot do its job must show up in the cron log");
+      assert.match(r.stdout, /busy-tools=MISSING/);
+      assert.match(
+        r.stdout,
+        /cannot prove idle \(lsof missing — apt install lsof\), kept: .*e2e-build\.tar\.gz/
+      );
+      for (const p of [f.staleTar, f.staleBuild, f.staleUpgrade, f.fresh, f.unrelated]) {
+        assert.ok(existsSync(p), `must not delete ${p} when idleness cannot be proven`);
+      }
+    } finally {
+      rmSync(f.base, { recursive: true, force: true });
+    }
+  });
+
+  it("--dry-run names what it WOULD remove and removes nothing", (t) => {
+    if (!HAVE_BUSY_TOOLS) return t.skip("lsof absent on this box — sweep branch covered in CI");
+    const f = fixture();
+    try {
+      const r = run(["--dry-run"], f.base);
+      assert.equal(r.status, 0, r.stderr + r.stdout);
+      assert.match(r.stdout, /busy-tools=ok/);
+      for (const p of [f.staleTar, f.staleBuild, f.staleUpgrade]) {
+        assert.match(
+          r.stdout,
+          new RegExp(`would remove \\(3h\\+\\): ${p.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")}`)
+        );
+        assert.doesNotMatch(
+          r.stdout,
+          new RegExp(`removed \\(3h\\+\\): ${p.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")}`),
+          "dry-run must never claim it removed something"
+        );
+        assert.ok(existsSync(p), `dry-run must not delete ${p}`);
+      }
+      assert.doesNotMatch(
+        r.stdout,
+        /omniroute-batch-api-fresh/,
+        "a fresh dir is never a candidate"
+      );
+      assert.doesNotMatch(r.stdout, /somebody-elses\.log/, "only names our tooling creates");
+      assert.match(r.stdout, /zombie builds: 0/);
+      assert.match(r.stdout, /done status=0/);
+    } finally {
+      rmSync(f.base, { recursive: true, force: true });
+    }
+  });
+
+  it("for real: sweeps the three stale artefacts, keeps the fresh one and the stranger", (t) => {
+    if (!HAVE_BUSY_TOOLS) return t.skip("lsof absent on this box — sweep branch covered in CI");
+    const f = fixture();
+    try {
+      const r = run([], f.base);
+      assert.equal(r.status, 0, r.stderr + r.stdout);
+      assert.ok(!existsSync(f.staleTar), "stale e2e-build.tar.gz must go (it is RAM on tmpfs)");
+      assert.ok(!existsSync(f.staleBuild), "stale next-build dir must go");
+      assert.ok(!existsSync(f.staleUpgrade), "stale install-upgrade dir must go");
+      assert.ok(existsSync(f.fresh), "a fresh dir must survive");
+      assert.ok(existsSync(f.unrelated), "files we did not create must survive even when old");
+    } finally {
+      rmSync(f.base, { recursive: true, force: true });
+    }
+  });
+
+  it("tmpfs fuse is shorter than the disk fuse (RAM vs disk), both overridable", () => {
+    const f = fixture();
+    try {
+      // With a 6h tmpfs fuse the 5h-old artefacts are NOT stale yet.
+      const r = run(["--dry-run"], f.base, { TMPFS_MAX_AGE_HOURS: "6" });
+      assert.doesNotMatch(
+        r.stdout,
+        /would remove|removed \(|cannot prove idle/,
+        "nothing is stale under a 6h fuse, so no candidate is even examined"
+      );
+      const body = readFileSync(SCRIPT, "utf8");
+      assert.match(body, /TMPFS_MAX_AGE_HOURS:-3\}/, "tmpfs default must stay short — it is RAM");
+      assert.match(body, /WORK_TEMP_MAX_AGE_HOURS:-24\}/);
+    } finally {
+      rmSync(f.base, { recursive: true, force: true });
+    }
+  });
+
+  it("alerts (exit 1) on disk and memory pressure thresholds without touching files", () => {
+    const f = fixture();
+    try {
+      writeFileSync(
+        path.join(f.base, "psi"),
+        "some avg10=0.00 avg60=0.00 avg300=0.00 total=1\nfull avg10=0.00 avg60=23.50 avg300=9.00 total=1\n"
+      );
+      const r = run(["--dry-run"], f.base, {
+        JANITOR_PSI_FILE: path.join(f.base, "psi"),
+        DISK_ALERT_PCT: "0",
+      });
+      assert.equal(r.status, 1, "attention needed must be exit 1 for the cron log");
+      assert.match(r.stdout, /MEMORY PRESSURE psi full\/avg60=23\.50%/);
+      assert.ok(existsSync(f.fresh) && existsSync(f.unrelated));
+      assert.match(r.stdout, /ROOT DISK \d+% >= 0%/);
+      assert.ok(existsSync(f.staleTar), "alerting never deletes");
+    } finally {
+      rmSync(f.base, { recursive: true, force: true });
+    }
+  });
+
+  it("rejects unknown arguments instead of silently running", () => {
+    const r = run(["--yolo"], os.tmpdir());
+    assert.equal(r.status, 2);
+  });
+});

tests/unit/skills-injection.test.ts

@@ -81,7 +81,7 @@ test("injectSkills renders enabled tools in provider-specific shapes", async ()
     function: {
       name: "omr_skill_c2VhcmNoQDEuMC4w", // encodedName("search@1.0.0")
       description: "search the web",
-      parameters: { type: "object", properties: { query: "string" } },
+      parameters: { type: "object", properties: { query: { type: "string" } } },
     },
   });
   assert.equal(decodeSkillToolName("omr_skill_c2VhcmNoQDEuMC4w"), "search@1.0.0");
@@ -90,14 +90,14 @@ test("injectSkills renders enabled tools in provider-specific shapes", async ()
     {
       name: "omr_skill_c2VhcmNoQDEuMC4w",
       description: "search the web",
-      input_schema: { type: "object", properties: { query: "string" } },
+      input_schema: { type: "object", properties: { query: { type: "string" } } },
     },
   ]);
   assert.deepEqual(geminiTools, [
     {
       name: "omr_skill_c2VhcmNoQDEuMC4w",
       description: "search the web",
-      parameters: { type: "object", properties: { query: "string" } },
+      parameters: { type: "object", properties: { query: { type: "string" } } },
     },
   ]);
   assert.deepEqual(fallbackTools, [openaiTools[0]]);
@@ -219,7 +219,7 @@ test("injectSkills auto mode matches message/context semantics and applies score
     function: {
       name: encodedName("issueSearch@1.0.0"),
       description: "search github issues and pull requests",
-      parameters: { type: "object", properties: { query: "string" } },
+      parameters: { type: "object", properties: { query: { type: "string" } } },
     },
   });
 });
@@ -338,3 +338,60 @@ test("injectSkills auto mode limits selected auto skills and keeps on-mode skill
   assert.equal(names.includes("alwaysOnUtility@1.0.0"), true);
   assert.equal(names.filter((name) => name.startsWith("searchSkill")).length, 5);
 });
+
+/**
+ * Regression for #11856 — injected skill tools carried a malformed JSON Schema.
+ *
+ * Skills may declare their input in shorthand (`{ "content": "string" }`).
+ * normalizeInputSchema() wrapped that bare property map as
+ * `{ type: "object", properties: { content: "string" } }` without expanding the
+ * shorthand values — and `"string"` is not a JSON Schema object. Zhipu GLM
+ * behind the Console Go tier validates tool schemas strictly and rejected the
+ * whole request with `[1210] Invalid API parameter`, giving a 100% failure rate
+ * on that provider regardless of request content or credentials. Most other
+ * providers tolerate the malformed schema, which is why it surfaced late.
+ *
+ * SkillSchema is `z.record(z.string(), z.unknown())`, so shorthand values pass
+ * validation from every skill source — the skills API, the GitHub collector and
+ * the skillssh marketplace alike.
+ */
+test("#11856 injectSkills expands shorthand property types into valid JSON Schema", async () => {
+  await skillRegistry.register({
+    name: "generation",
+    version: "1.0.0",
+    description: "generate content",
+    schema: {
+      input: {
+        content: "string",
+        count: "number",
+        // already-expanded entries must survive untouched
+        options: { type: "object", properties: { tone: { type: "string" } } },
+      },
+      output: { result: "string" },
+    },
+    handler: "generation-handler",
+    enabled: true,
+    apiKeyId: "key-11856",
+  });
+
+  const expected = {
+    type: "object",
+    properties: {
+      content: { type: "string" },
+      count: { type: "number" },
+      options: { type: "object", properties: { tone: { type: "string" } } },
+    },
+  };
+
+  const openaiTools = injectSkills({ provider: "openai", apiKeyId: "key-11856" });
+  assert.deepEqual(
+    (openaiTools[0] as { function: { parameters: unknown } }).function.parameters,
+    expected
+  );
+
+  const claudeTools = injectSkills({ provider: "anthropic", apiKeyId: "key-11856" });
+  assert.deepEqual((claudeTools[0] as { input_schema: unknown }).input_schema, expected);
+
+  const geminiTools = injectSkills({ provider: "google", apiKeyId: "key-11856" });
+  assert.deepEqual((geminiTools[0] as { parameters: unknown }).parameters, expected);
+});

tests/unit/stream-markdown-token-boundary.test.ts

@@ -203,7 +203,10 @@ test("OpenAI to Claude: finish flushes a fully-held boundary before message stop
     },
     state
   );
-  const result = flatten([chunk1, chunk2]);
+  // End-of-stream flush (see dd35750e5f): a finish chunk without usage is deferred
+  // until production's null flush, so mirror it before asserting the terminal events.
+  const chunk3 = openaiToClaudeResponse(null, state);
+  const result = flatten([chunk1, chunk2, chunk3]);
 
   assert.deepEqual(getTextDeltas(result), ["`"]);
   assert.equal(state._markdownBuffer, "");
@@ -251,7 +254,10 @@ test("OpenAI to Claude: tool call flushes a fully-held boundary before tool use"
     },
     state
   );
-  const result = flatten([chunk1, chunk2]);
+  // End-of-stream flush (see dd35750e5f): a finish chunk without usage is deferred
+  // until production's null flush, so mirror it before asserting the terminal events.
+  const chunk3 = openaiToClaudeResponse(null, state);
+  const result = flatten([chunk1, chunk2, chunk3]);
   const contentEvents = result.filter((event) =>
     String((event as Record<string, unknown>).type).startsWith("content_block_")
   );

tests/unit/strip-store-responses.test.ts

@@ -0,0 +1,120 @@
+import assert from "node:assert/strict";
+import { createServer } from "node:http";
+import test from "node:test";
+
+import { DefaultExecutor } from "../../open-sse/executors/default.ts";
+import { stripStore } from "../../open-sse/handlers/chatCore/agentRouterProtocol.ts";
+import { FORMATS } from "../../open-sse/translator/formats.ts";
+
+const COMPATIBLE_PROVIDER = "openai-compatible-responses-test";
+
+test("stripStore forces store=false for stateless OpenAI-compatible Responses requests", () => {
+  for (const initialStore of [undefined, false, true]) {
+    const body: Record<string, unknown> = {};
+    if (initialStore !== undefined) body.store = initialStore;
+
+    stripStore(body, COMPATIBLE_PROVIDER, FORMATS.OPENAI_RESPONSES, {});
+
+    assert.equal(body.store, false);
+  }
+});
+
+test("stripStore preserves client store values for opted-in OpenAI-compatible Responses requests", () => {
+  for (const initialStore of [false, true]) {
+    const body: Record<string, unknown> = { store: initialStore };
+
+    stripStore(body, COMPATIBLE_PROVIDER, FORMATS.OPENAI_RESPONSES, {
+      openaiStoreEnabled: true,
+    });
+
+    assert.equal(body.store, initialStore);
+  }
+
+  const omitted: Record<string, unknown> = {};
+  stripStore(omitted, COMPATIBLE_PROVIDER, FORMATS.OPENAI_RESPONSES, {
+    openaiStoreEnabled: true,
+  });
+  assert.equal("store" in omitted, false);
+});
+
+test("stripStore keeps existing OpenAI and AgentRouter behavior", () => {
+  const cases = [
+    { provider: "openai", targetFormat: FORMATS.OPENAI, expected: true },
+    { provider: "openai", targetFormat: FORMATS.OPENAI_RESPONSES, expected: true },
+    { provider: "agentrouter", targetFormat: FORMATS.OPENAI_RESPONSES, expected: true },
+    { provider: "agentrouter", targetFormat: FORMATS.OPENAI, expected: false },
+  ];
+
+  for (const { provider, targetFormat, expected } of cases) {
+    const body: Record<string, unknown> = { store: true };
+    stripStore(body, provider, targetFormat, {});
+    assert.equal("store" in body, expected, `${provider}/${targetFormat}`);
+  }
+});
+
+test("stripStore removes store outside OpenAI-compatible Responses targets", () => {
+  const cases = [
+    { provider: COMPATIBLE_PROVIDER, targetFormat: FORMATS.OPENAI },
+    { provider: "anthropic", targetFormat: FORMATS.CLAUDE },
+  ];
+
+  for (const { provider, targetFormat } of cases) {
+    const body: Record<string, unknown> = { store: false };
+    stripStore(body, provider, targetFormat, { openaiStoreEnabled: true });
+    assert.equal("store" in body, false, `${provider}/${targetFormat}`);
+  }
+});
+
+test("DefaultExecutor never serializes native passthrough markers upstream", async () => {
+  let capturedBody: Record<string, unknown> | null = null;
+  const server = createServer((request, response) => {
+    let rawBody = "";
+    request.setEncoding("utf8");
+    request.on("data", (chunk) => {
+      rawBody += chunk;
+    });
+    request.on("end", () => {
+      capturedBody = JSON.parse(rawBody);
+      response.writeHead(200, { "Content-Type": "application/json" });
+      response.end(JSON.stringify({ id: "resp_test", object: "response", output: [] }));
+    });
+  });
+  await new Promise<void>((resolve) => server.listen(0, "127.0.0.1", resolve));
+  const address = server.address();
+  assert.ok(address && typeof address !== "string");
+
+  try {
+    const executor = new DefaultExecutor(COMPATIBLE_PROVIDER);
+    await executor.execute({
+      model: "gpt-5.6-test",
+      body: {
+        model: "gpt-5.6-test",
+        input: "hi",
+        store: false,
+        _nativeOpenAICompatibleResponsesPassthrough: true,
+        _nativeCodexPassthrough: true,
+        _nativeXaiResponsesPassthrough: true,
+        _omnirouteResponsesStore: false,
+      },
+      stream: false,
+      credentials: {
+        apiKey: "test-key",
+        providerSpecificData: {
+          apiType: "responses",
+          baseUrl: `http://127.0.0.1:${address.port}/v1`,
+        },
+      },
+    });
+  } finally {
+    await new Promise<void>((resolve, reject) =>
+      server.close((error) => (error ? reject(error) : resolve()))
+    );
+  }
+
+  assert.ok(capturedBody);
+  assert.equal(capturedBody.store, false);
+  assert.equal(capturedBody._nativeOpenAICompatibleResponsesPassthrough, undefined);
+  assert.equal(capturedBody._nativeCodexPassthrough, undefined);
+  assert.equal(capturedBody._nativeXaiResponsesPassthrough, undefined);
+  assert.equal(capturedBody._omnirouteResponsesStore, undefined);
+});

tests/unit/token-health-check-webcookie.test.ts

@@ -29,7 +29,7 @@ function baseParams(over: Partial<ProbeParams> = {}): ProbeParams {
 describe("web-cookie health probe (#11488)", () => {
   it("candidate detection matches catalogued cookie providers only", () => {
     assert.equal(isWebCookieHealthProbeCandidate("claude-web"), true);
-    assert.equal(isWebCookieHealthProbeCandidate("chatgpt-web"), true);
+    assert.equal(isWebCookieHealthProbeCandidate("chatgpt-web-codex"), true);
     assert.equal(isWebCookieHealthProbeCandidate("openai"), false);
     assert.equal(isWebCookieHealthProbeCandidate(undefined), false);
     assert.equal(isWebCookieHealthProbeCandidate(""), false);
@@ -210,7 +210,7 @@ describe("web-cookie health probe (#11488)", () => {
       baseParams({
         conn: {
           id: "c1",
-          provider: "qwen-web",
+          provider: "grok-web",
           apiKey: "",
           providerSpecificData: { cookie: "token=abc" },
         },

tests/unit/translator-tool-call-shim.test.ts

@@ -1,12 +1,10 @@
 import test from "node:test";
 import assert from "node:assert/strict";
 
-const { applyToolCallShimToBuffer, hasToolCallShim, __test } = await import(
-  "../../open-sse/translator/helpers/toolCallShim.ts"
-);
-const { openaiToClaudeResponse } = await import(
-  "../../open-sse/translator/response/openai-to-claude.ts"
-);
+const { applyToolCallShimToBuffer, hasToolCallShim, __test } =
+  await import("../../open-sse/translator/helpers/toolCallShim.ts");
+const { openaiToClaudeResponse } =
+  await import("../../open-sse/translator/response/openai-to-claude.ts");
 
 const { coerceToArray } = __test as { coerceToArray: (v: unknown) => unknown[] };
 
@@ -120,30 +118,21 @@ test("applyToolCallShimToBuffer: Read coerces numeric-string limit/offset", () =
 
 test("applyToolCallShimToBuffer: Read strips pages for non-PDF files", () => {
   const out = JSON.parse(
-    applyToolCallShimToBuffer(
-      "Read",
-      JSON.stringify({ file_path: "/etc/hosts", pages: "1-3" })
-    )
+    applyToolCallShimToBuffer("Read", JSON.stringify({ file_path: "/etc/hosts", pages: "1-3" }))
   );
   assert.equal("pages" in out, false);
 });
 
 test("applyToolCallShimToBuffer: Read strips malformed pages even on PDFs", () => {
   const out = JSON.parse(
-    applyToolCallShimToBuffer(
-      "Read",
-      JSON.stringify({ file_path: "/tmp/doc.pdf", pages: "abc" })
-    )
+    applyToolCallShimToBuffer("Read", JSON.stringify({ file_path: "/tmp/doc.pdf", pages: "abc" }))
   );
   assert.equal("pages" in out, false);
 });
 
 test("applyToolCallShimToBuffer: Read accepts a single page on PDFs", () => {
   const out = JSON.parse(
-    applyToolCallShimToBuffer(
-      "Read",
-      JSON.stringify({ file_path: "/tmp/doc.PDF", pages: "7" })
-    )
+    applyToolCallShimToBuffer("Read", JSON.stringify({ file_path: "/tmp/doc.PDF", pages: "7" }))
   );
   assert.equal(out.pages, "7");
 });
@@ -255,6 +244,12 @@ function streamChunks(chunks: any[], state: any): any[] {
     const out = openaiToClaudeResponse(c, state);
     if (out) all.push(...out);
   }
+  // End-of-stream flush: production calls the translator once more with `null`
+  // when the upstream stream closes (open-sse/utils/stream.ts flush →
+  // translateResponse(..., null, state)). Since dd35750e5f a finish chunk that
+  // carries no usage is deferred until that flush, so the driver must mirror it.
+  const flushed = openaiToClaudeResponse(null, state);
+  if (flushed) all.push(...flushed);
   return all;
 }
 

tests/unit/ui/ProviderIcon-icon-url.test.tsx

@@ -243,3 +243,24 @@ describe("ProviderIcon — unresolved local asset provenance", () => {
     }
   );
 });
+
+// #11853 follow-up: getLobeProviderIcon() itself is already guarded by #11880's
+// Object.hasOwn() checks (see lobe-provider-icons-prototype-collision-11853.test.ts).
+// This covers the three *other* plain-object lookups ProviderIcon.tsx does on its own
+// (PROVIDER_ICON_ALIASES, LOCAL_SVG_ALIASES, THEMED_SVGS) — none of which #11880 touched —
+// which resolved the same inherited-property ids through the prototype chain before
+// falling through to the thesvg.org unknown-provider CDN path.
+describe("ProviderIcon — inherited object property ids", () => {
+  it.each(["constructor", "valueOf", "hasOwnProperty", "__proto__"])(
+    "renders provider id %s through the unknown-provider fallback",
+    (providerId) => {
+      const container = renderIcon({ providerId });
+      const img = container.querySelector("img");
+
+      expect(img).not.toBeNull();
+      expect(img?.getAttribute("src")).toBe(
+        `https://thesvg.org/icons/${providerId.toLowerCase()}/default.svg`
+      );
+    }
+  );
+});

tests/unit/web-session-contract.test.ts

@@ -47,7 +47,7 @@ test("web-session contract preserves representative token and cookie semantics",
   assert.equal(providers.get("deepseek-web")?.credential.kind, "token");
   assert.equal(providers.get("zai-web")?.credential.kind, "token");
   assert.equal(providers.get("gemini-web")?.credential.kind, "cookie");
-  assert.equal(providers.get("qwen-web")?.credential.kind, "cookie");
+  assert.equal(providers.get("perplexity-web")?.credential.kind, "cookie");
 
   assert.ok(
     providers

tests/unit/xai-translators.test.ts

@@ -60,19 +60,27 @@ test("applyThinking: honors xAI-native reasoning.effort verbatim", () => {
   assert.equal((out as Record<string, unknown>).foo, 1);
 });
 
-test("normalizeXaiReasoningEffort: downgrades max/xhigh to xAI-supported high", () => {
+test("normalizeXaiReasoningEffort: downgrades max, passes xhigh through (#11816)", () => {
+  // "max" is not an xAI tier -> still folded onto "high".
   assert.equal(normalizeXaiReasoningEffort("max"), "high");
-  assert.equal(normalizeXaiReasoningEffort("xhigh"), "high");
+  // "xhigh" is a real xAI tier on grok-4.6+; xAI itself degrades it to "high"
+  // on older models, so forwarding it verbatim is always safe.
+  assert.equal(normalizeXaiReasoningEffort("xhigh"), "xhigh");
+  assert.equal(normalizeXaiReasoningEffort("XHIGH"), "xhigh");
   assert.equal(normalizeXaiReasoningEffort("HIGH"), "high");
   assert.equal(normalizeXaiReasoningEffort("ultra"), undefined);
 });
 
-test("applyThinking: normalizes xAI-native max/xhigh to high", () => {
+test("applyThinking: folds max to high, keeps xhigh intact (#11816)", () => {
   const maxOut = applyThinking({ reasoning: { effort: "max", summary: "auto" } });
   assert.deepStrictEqual(maxOut.reasoning, { effort: "high", summary: "auto" });
 
   const xhighOut = applyThinking({ reasoning: { effort: "xhigh" } });
-  assert.deepStrictEqual(xhighOut.reasoning, { effort: "high" });
+  assert.deepStrictEqual(xhighOut.reasoning, { effort: "xhigh" });
+
+  const chatOut = applyThinking({ reasoning_effort: "xhigh" });
+  assert.deepStrictEqual(chatOut.reasoning, { effort: "xhigh" });
+  assert.equal(chatOut.reasoning_effort, undefined);
 });
 
 test("applyThinking: rewrites OpenAI Chat reasoning_effort into reasoning.effort", () => {