fix(release): attach the SBOM to the GitHub Release on dispatch publishes too (#12020 )

The step was gated on github.event_name == 'release'. v3.8.50's package shipped through a workflow_dispatch (the staged publish, 11 attempts) and the step was skipped, so the GitHub Release carried no SBOM — it was attached by hand from the run's sbom-npm artifact (5.0 MB, 1,886 components). Now it attaches on release or workflow_dispatch whenever a release for the published tag exists, and says so when it does not (the workflow artifact remains the durable copy either way). actionlint and prettier clean; npm-publish-artifact-provenance and check-workflows-provenance-runner suites pass.
fix(release): resync the electron lockfile and let a dispatch build from a repaired ref (#11982 )
2026-08-29 18:42:17 +03:00 · 2026-08-29 10:02:51 -03:00 · 2026-08-29 09:05:34 -03:00 · 2026-08-29 06:47:17 -03:00 · 2026-08-29 03:15:46 -03:00 · 2026-08-28 19:01:54 -03:00
89 changed files with 2275 additions and 245 deletions
--- a/.env.example
+++ b/.env.example
@@ -1023,6 +1023,13 @@ PROVIDER_LIMITS_SYNC_SPACING_MS=1500
 # to disable the check. Used by: src/lib/db/migrationRunner.ts.  Default: 50.
 #OMNIROUTE_MAX_PENDING_MIGRATIONS=50

+# Working directory for the check:install-upgrade release gate. It builds two ~3 GB
+# install trees plus a ~275 MB tarball, so it needs roughly 12 GB — more than the
+# 12 GB RAM-backed tmpfs that /tmp is on the self-hosted runner, where it exhausted
+# the tmpfs and npm silently truncated the package. Defaults to <repo>/.install-upgrade
+# on real disk. Used by: scripts/check/check-install-upgrade.mjs.  Default: <repo>/.install-upgrade.
+#OMNIROUTE_INSTALL_UPGRADE_WORKDIR=/var/tmp/omniroute-install-upgrade
+
 # Trust user-managed RTK project filter rules without strict signature checks.
 # Used by: open-sse/services/compression/engines/rtk/filterLoader.ts.  Default: 0.
 #OMNIROUTE_RTK_TRUST_PROJECT_FILTERS=0
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -606,13 +606,24 @@ 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: ${{ github.event_name != 'pull_request' || (needs.changes.outputs.code == 'true' && github.event.pull_request.draft == false) }}
    steps:
      - uses: actions/checkout@v7
@@ -646,14 +657,14 @@ jobs:
        # Keep standalone/node_modules intact: package/electron jobs consume the
        # Next-traced standalone tree and must not replace it with root node_modules.
        run: |
-          tar -czf /tmp/e2e-build.tar.gz \
+          tar -czf "$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:
@@ -676,10 +687,14 @@ 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
@@ -767,10 +782,14 @@ 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
@@ -957,7 +976,11 @@ jobs:
    # 10min was sized before #7114 added the lcov reporter (Codecov/Sonar need it);
    # merging 8 shard JSONs + text+json+lcov now takes ~10-12min — three consecutive
    # release-tip runs died at exactly 10m as job-timeout "cancelled" (2026-07-15/16).
-    timeout-minutes: 20
+    # 30, not 20 (2026-08-29): the informational Codecov upload below hung for the rest of
+    # the budget on two consecutive main runs (33207760653, 33215115341); the job ended
+    # `cancelled` and dragged the whole run's conclusion to `cancelled` although every
+    # blocking job was green. The upload step now has its own ceiling; this is headroom.
+    timeout-minutes: 30
    needs: test-unit
    if: ${{ !cancelled() && needs.test-unit.result == 'success' && !contains(github.event.pull_request.labels.*.name, 'hotfix') }}
    env:
@@ -1036,6 +1059,10 @@ jobs:
      # (if-no-files-found: warn) — Sonar consumes the same file.
      - name: Upload coverage to Codecov (informational)
        if: always()
+        # Informational means informational: its own ceiling and continue-on-error, so a
+        # stalled upload can neither eat the job's budget nor turn a green job cancelled.
+        timeout-minutes: 5
+        continue-on-error: true
        uses: codecov/codecov-action@fb8b3582c8e4def4969c97caa2f19720cb33a72f # v7.0.0
        with:
          files: coverage/lcov.info
@@ -1230,10 +1257,14 @@ 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
--- a/.github/workflows/docker-publish.yml
+++ b/.github/workflows/docker-publish.yml
@@ -185,6 +185,13 @@ jobs:

      - name: Build and push BUN base platform image by digest
        id: build-bun-base
+        # Bun is a best-effort compatibility target, not a supported runtime
+        # (AGENTS.md -> Environment). Its `bun run build` has been OOM-killing on
+        # both arches; letting that sink the whole publish means the SUPPORTED
+        # runner-base / runner-web images never reach the registry either. The
+        # image is still built and pushed whenever it succeeds — only its power to
+        # block the release is removed.
+        continue-on-error: true
        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7
        with:
          context: .
@@ -203,6 +210,13 @@ jobs:

      - name: Build and push BUN web platform image by digest
        id: build-bun-web
+        # Bun is a best-effort compatibility target, not a supported runtime
+        # (AGENTS.md -> Environment). Its `bun run build` has been OOM-killing on
+        # both arches; letting that sink the whole publish means the SUPPORTED
+        # runner-base / runner-web images never reach the registry either. The
+        # image is still built and pushed whenever it succeeds — only its power to
+        # block the release is removed.
+        continue-on-error: true
        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7
        with:
          context: .
@@ -230,8 +244,15 @@ jobs:
          mkdir -p /tmp/digests/base /tmp/digests/web /tmp/digests/bun-base /tmp/digests/bun-web
          touch "/tmp/digests/base/${DIGEST_BASE#sha256:}"
          touch "/tmp/digests/web/${DIGEST_WEB#sha256:}"
-          touch "/tmp/digests/bun-base/${DIGEST_BUN_BASE#sha256:}"
-          touch "/tmp/digests/bun-web/${DIGEST_BUN_WEB#sha256:}"
+          # Empty when the (non-blocking) bun build produced no image. `if` blocks,
+          # not `[ -n ] && touch`: under `set -e` a failing AND-list aborts the step,
+          # which is precisely the case being handled here.
+          if [ -n "$DIGEST_BUN_BASE" ]; then
+            touch "/tmp/digests/bun-base/${DIGEST_BUN_BASE#sha256:}"
+          fi
+          if [ -n "$DIGEST_BUN_WEB" ]; then
+            touch "/tmp/digests/bun-web/${DIGEST_BUN_WEB#sha256:}"
+          fi

      - name: Upload base digests
        uses: actions/upload-artifact@v7
@@ -254,7 +275,11 @@ jobs:
        with:
          name: digests-bun-base-${{ matrix.arch }}
          path: /tmp/digests/bun-base/*
-          if-no-files-found: error
+          # `ignore`, not `error`: the bun build is non-blocking, so an absent
+          # digest is the expected outcome of a failed/skipped bun image — the
+          # manifest step already treats these tags as optional. Leaving `error`
+          # here just relocates the blocker from the manifest to the upload.
+          if-no-files-found: ignore
          retention-days: 1

      - name: Upload bun-web digests
@@ -262,7 +287,11 @@ jobs:
        with:
          name: digests-bun-web-${{ matrix.arch }}
          path: /tmp/digests/bun-web/*
-          if-no-files-found: error
+          # `ignore`, not `error`: the bun build is non-blocking, so an absent
+          # digest is the expected outcome of a failed/skipped bun image — the
+          # manifest step already treats these tags as optional. Leaving `error`
+          # here just relocates the blocker from the manifest to the upload.
+          if-no-files-found: ignore
          retention-days: 1

  merge:
@@ -320,6 +349,9 @@ jobs:
          merge-multiple: true

      - name: Download bun-base digests
+        # Non-blocking: the bun image is best-effort, so its artifact may not
+        # exist at all. The manifest step treats these tags as optional.
+        continue-on-error: true
        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8
        with:
          pattern: digests-bun-base-*
@@ -327,6 +359,9 @@ jobs:
          merge-multiple: true

      - name: Download bun-web digests
+        # Non-blocking: the bun image is best-effort, so its artifact may not
+        # exist at all. The manifest step treats these tags as optional.
+        continue-on-error: true
        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8
        with:
          pattern: digests-bun-web-*
@@ -338,7 +373,7 @@ jobs:
          set -euo pipefail

          create_manifest() {
-            local image="$1" suffix="$2" dir="$3"
+            local image="$1" suffix="$2" dir="$3" optional="${4:-}"
            local tags=(-t "${image}:${VERSION}${suffix}")
            if [ "$PROMOTE_LATEST" = "true" ]; then
              tags+=(-t "${image}:latest${suffix}")
@@ -348,6 +383,10 @@ jobs:
              refs+=("${image}@sha256:$(basename "$digest_file")")
            done < <(find "$dir" -type f | sort)
            if [ "${#refs[@]}" -eq 0 ]; then
+              if [ -n "$optional" ]; then
+                echo "::warning::No image digests in $dir — skipping optional tag ${image}:${VERSION}${suffix}" >&2
+                return 0
+              fi
              echo "No image digests in $dir" >&2
              exit 1
            fi
@@ -356,15 +395,15 @@ jobs:

          create_manifest "${IMAGE_NAME}" "" /tmp/digests/base
          create_manifest "${IMAGE_NAME}" "-web" /tmp/digests/web
-          create_manifest "${IMAGE_NAME}" "-bun" /tmp/digests/bun-base
-          create_manifest "${IMAGE_NAME}" "-web-bun" /tmp/digests/bun-web
+          create_manifest "${IMAGE_NAME}" "-bun" /tmp/digests/bun-base optional
+          create_manifest "${IMAGE_NAME}" "-web-bun" /tmp/digests/bun-web optional

      - name: Create GHCR manifest
        run: |
          set -euo pipefail

          create_manifest() {
-            local image="$1" suffix="$2" dir="$3"
+            local image="$1" suffix="$2" dir="$3" optional="${4:-}"
            local tags=(-t "${image}:${VERSION}${suffix}")
            if [ "$PROMOTE_LATEST" = "true" ]; then
              tags+=(-t "${image}:latest${suffix}")
@@ -374,6 +413,10 @@ jobs:
              refs+=("${image}@sha256:$(basename "$digest_file")")
            done < <(find "$dir" -type f | sort)
            if [ "${#refs[@]}" -eq 0 ]; then
+              if [ -n "$optional" ]; then
+                echo "::warning::No image digests in $dir — skipping optional tag ${image}:${VERSION}${suffix}" >&2
+                return 0
+              fi
              echo "No image digests in $dir" >&2
              exit 1
            fi
@@ -382,8 +425,8 @@ jobs:

          create_manifest "${GHCR_IMAGE_NAME}" "" /tmp/digests/base
          create_manifest "${GHCR_IMAGE_NAME}" "-web" /tmp/digests/web
-          create_manifest "${GHCR_IMAGE_NAME}" "-bun" /tmp/digests/bun-base
-          create_manifest "${GHCR_IMAGE_NAME}" "-web-bun" /tmp/digests/bun-web
+          create_manifest "${GHCR_IMAGE_NAME}" "-bun" /tmp/digests/bun-base optional
+          create_manifest "${GHCR_IMAGE_NAME}" "-web-bun" /tmp/digests/bun-web optional

      - name: Inspect image
        if: needs.prepare.outputs.version != 'main'
--- a/.github/workflows/electron-release.yml
+++ b/.github/workflows/electron-release.yml
@@ -10,6 +10,16 @@ on:
        description: "Release version (e.g., v1.6.8)"
        required: true
        type: string
+      publish_npm:
+        description: "Also run the npm publish leg (turn off when re-attaching desktop assets to a release whose npm package already shipped)"
+        required: false
+        default: true
+        type: boolean
+      build_ref:
+        description: "Git ref to BUILD from (default: the version tag). Set to a branch when the tag itself cannot build — e.g. a lockfile that was already broken when it was cut — and the assets must come from the repaired line"
+        required: false
+        default: ""
+        type: string

 # Least-privilege default: read-only at the top level; each job grants the writes it
 # needs (build/release upload assets, publish-npm forwards npm provenance / packages
@@ -76,6 +86,9 @@ jobs:
      - uses: actions/checkout@v7
        with:
          persist-credentials: false
+          # workflow_dispatch: build the tag being (re)built, not the dispatching branch. On a
+          # tag push this resolves to the same commit.
+          ref: ${{ inputs.build_ref || needs.validate.outputs.version }}
      - name: Setup Node
        uses: actions/setup-node@v7
        with:
@@ -161,6 +174,9 @@ jobs:
      - uses: actions/checkout@v7
        with:
          persist-credentials: false
+          # workflow_dispatch: build the tag being (re)built, not the dispatching branch. On a
+          # tag push this resolves to the same commit.
+          ref: ${{ inputs.build_ref || needs.validate.outputs.version }}
      - name: Setup Node
        uses: actions/setup-node@v7
        with:
@@ -347,6 +363,8 @@ jobs:
        with:
          persist-credentials: false
          fetch-depth: 0
+          # Source archives + SBOM come from the tag being released, not the dispatching branch.
+          ref: ${{ inputs.build_ref || needs.validate.outputs.version }}

      # `merge-multiple` is deliberately OFF. It resolves same-name collisions by ARRIVAL
      # ORDER, and the two macOS jobs each emit their own `latest-mac.yml` listing only their
@@ -404,7 +422,10 @@ jobs:
          tag_name: ${{ needs.validate.outputs.version }}
          draft: false
          prerelease: false
-          generate_release_notes: true
+          # Only on the tag push. A re-attach dispatch runs against a release whose curated
+          # notes already exist, and `true` APPENDS GitHub's auto-generated "What's Changed"
+          # block to them (v3.8.50, run 33238093090: +1,416 chars on a 121 KB body).
+          generate_release_notes: ${{ github.event_name != 'workflow_dispatch' }}
          fail_on_unmatched_files: false
          files: |
            release-assets/*.dmg
@@ -462,11 +483,20 @@ jobs:
  publish-npm:
    name: Publish to npm
    needs: [validate, release]
+    # A re-dispatch that only re-attaches desktop assets must not publish the npm package again.
+    if: ${{ github.event_name != 'workflow_dispatch' || inputs.publish_npm }}
    permissions:
      # Must be `write`, not `read`: this job calls the reusable npm-publish.yml whose
      # `publish` job needs `contents: write` (gh release upload — attach the SBOM, #3874).
      # A reusable workflow's job cannot request more permission than the caller grants,
      # so a `read` here makes GitHub reject the run at startup (startup_failure).
+      #
+      # `actions: read` for the same reason: the called `publish` job downloads the next-build
+      # artefact and requests it. v3.8.50 (run 33005490476) died at startup with "The nested
+      # job 'publish' is requesting 'actions: read', but is only allowed 'actions: none'" — and
+      # because `release` lives in this same workflow, the tag shipped with ZERO assets. Keep
+      # this block a superset of every job's permissions in npm-publish.yml.
+      actions: read
      contents: write
      id-token: write # npm provenance (forwarded to the reusable workflow)
      packages: write # publish to npm.pkg.github.com
--- a/.github/workflows/nightly-release-green.yml
+++ b/.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
--- a/.github/workflows/npm-publish.yml
+++ b/.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:
@@ -62,11 +63,15 @@ 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 }}
+      skip: ${{ steps.resolve.outputs.skip }}
    permissions:
      actions: read # find + download the CI run's next-build artifact for this SHA
      contents: write # gh release upload (attach SBOM to the GitHub Release)
-      id-token: write # npm provenance
+      id-token: write # npm provenance (GitHub Packages step)
      packages: write # publish to npm.pkg.github.com
    steps:
      - name: Checkout
@@ -200,8 +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="$candidate"
              break
            fi
@@ -211,8 +219,8 @@ jobs:
            echo "::notice::none of the candidate runs still carries next-build (1-day retention) — falling back to a full build"
            exit 0
          fi
-          tar -xzf /tmp/next-build/e2e-build.tar.gz -C .
-          rm -rf /tmp/next-build
+          tar -xzf "$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
@@ -226,6 +234,28 @@ jobs:
          JWT_SECRET: ci-build-secret-with-sufficient-length-for-validation
        run: npm run build:cli

+      # `build:cli` assembles dist/ but does NOT write dist/BUILD_SHA — only
+      # `build:release` does, by calling write-build-sha.mjs. The #10427 provenance
+      # guard inside check:pack-artifact rejects an artifact with no SHA (and rejects
+      # it even under OMNIROUTE_ALLOW_CANARY_BUILD=1: what cannot be identified cannot
+      # be vouched for). Without this step the build+validate pair in this job is
+      # structurally incompatible and fails 100% of the time — the same gap that was
+      # fixed in ci.yml's Package Artifact job.
+      - name: Stamp dist/BUILD_SHA for the provenance guard (#10427)
+        if: steps.resolve.outputs.skip != 'true'
+        env:
+          OMNIROUTE_BUILD_SHA: ${{ github.sha }}
+        run: |
+          export OMNIROUTE_BUILD_SHA="${OMNIROUTE_BUILD_SHA:0:7}"
+          node scripts/build/write-build-sha.mjs
+
+      # The guard checks ancestry against origin/main by default, which is correct
+      # here (a release tag is cut from main), but the ref has to exist locally for
+      # `git merge-base` to resolve it.
+      - name: Fetch main for the provenance probe
+        if: steps.resolve.outputs.skip != 'true'
+        run: git fetch --no-tags --depth=50 origin +refs/heads/main:refs/remotes/origin/main
+
      - name: Validate npm package artifact
        if: steps.resolve.outputs.skip != 'true'
        run: npm run check:pack-artifact
@@ -243,11 +273,20 @@ jobs:
          if-no-files-found: error

      - name: Attach SBOM to GitHub Release
-        if: steps.resolve.outputs.skip != 'true' && github.event_name == 'release'
+        # Not only on the `release` event: the v3.8.50 package shipped through a
+        # workflow_dispatch (staged publish, 11 attempts) and this step was skipped, so the
+        # GitHub Release carried no SBOM until it was attached by hand from the run's
+        # `sbom-npm` artifact. Attach whenever a release for the published tag exists.
+        if: steps.resolve.outputs.skip != 'true' && (github.event_name == 'release' || github.event_name == 'workflow_dispatch')
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-          TAG: ${{ github.ref_name }}
-        run: gh release upload "$TAG" sbom-npm.cdx.json --clobber
+          TAG: ${{ github.event_name == 'release' && github.ref_name || format('v{0}', inputs.version) }}
+        run: |
+          if ! gh release view "$TAG" --repo "$GITHUB_REPOSITORY" >/dev/null 2>&1; then
+            echo "::notice::no GitHub Release for $TAG yet — SBOM stays on the sbom-npm workflow artifact"
+            exit 0
+          fi
+          gh release upload "$TAG" sbom-npm.cdx.json --repo "$GITHUB_REPOSITORY" --clobber

      # WS1.2/WS1.3 (#7065 class): the artifact that is about to be published must
      # BOOT. build:cli already assembled dist/ above; this packs+installs+boots the
@@ -265,7 +304,12 @@ jobs:
      # a staged package that is never approved simply expires, with no `npm deprecate` needed.
      - name: Prove clean-install AND upgrade-over-previous both boot
        if: steps.resolve.outputs.skip != 'true'
-        timeout-minutes: 30
+        # 60, not 30. This gate was added in #8953 and the 2026-08-27 v3.8.50 publish
+        # was the FIRST run to ever reach it — every earlier attempt died upstream, so
+        # its budget had never been measured against a real run. It then blew the limit
+        # on its debut: `npm pack` alone took 24m37s, leaving 5 minutes for two installs
+        # and two boots. 30 was a guess; 60 is sized to the one measurement we have.
+        timeout-minutes: 60
        run: npm run check:install-upgrade

      # WS1.3 (D2, v3.8.49 plan): STAGED publishing by default — `npm stage publish`
@@ -287,41 +331,33 @@ jobs:
          fi
          npm --version

-      - name: Publish to npm (staged — owner approves with 2FA)
-        if: steps.resolve.outputs.skip != 'true' && (github.event_name != 'workflow_dispatch' || inputs.publish_mode != 'direct')
+      # The registry upload itself moved to the `stage-npm` job below: npm REFUSES
+      # `--provenance` from a self-hosted runner (422 "Unsupported GitHub Actions
+      # runner environment"), and the heavy verification above cannot move to a
+      # hosted one (16 GB is not enough for build:cli's next-build fallback — see
+      # this job's runs-on comment). So this job proves the bytes and hands them
+      # over; a tiny hosted job does the upload.
+      - name: Pack the verified tarball for the upload job
+        if: steps.resolve.outputs.skip != 'true'
        env:
          VERSION: ${{ steps.resolve.outputs.version }}
-          TAG: ${{ steps.resolve.outputs.tag }}
-          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
        run: |
          set -euo pipefail
-          # Always pass --tag explicitly. Defense in depth: even if VERSION is
-          # accidentally an older release, the historic tag will NOT claim `@latest`.
-          npm stage publish --provenance --access public --tag "$TAG"
-          {
-            echo "## 📦 omniroute@$VERSION STAGED (not yet installable)"
-            echo ""
-            echo "The exact bytes are parked on the registry. To release them:"
-            echo '```'
-            echo "npm stage list omniroute            # find the stage id"
-            echo "npm stage approve <id>              # owner 2FA — THE publish"
-            echo '```'
-            echo "To verify the staged bytes first: npm stage download <id> → run"
-            echo "scripts/check/check-pack-boot.mjs against them (see RELEASE_CHECKLIST)."
-            echo "To discard: npm stage reject <id>."
-          } >> "$GITHUB_STEP_SUMMARY"
-          echo "✅ Staged omniroute@$VERSION (dist-tag=$TAG) — awaiting owner 'npm stage approve'"
+          # --ignore-scripts: prepublishOnly would re-run build:cli-api && build:cli,
+          # rebuilding bytes this job has already built, validated and boot-smoked.
+          npm pack --ignore-scripts
+          TARBALL="omniroute-${VERSION}.tgz"
+          test -f "$TARBALL" || { echo "expected $TARBALL to exist after npm pack" >&2; ls -la ./*.tgz || true; exit 1; }
+          echo "packed $TARBALL ($(du -h "$TARBALL" | cut -f1))"

-      - name: Publish to npm (DIRECT — emergency fallback)
-        if: steps.resolve.outputs.skip != 'true' && github.event_name == 'workflow_dispatch' && inputs.publish_mode == 'direct'
-        env:
-          VERSION: ${{ steps.resolve.outputs.version }}
-          TAG: ${{ steps.resolve.outputs.tag }}
-          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
-        run: |
-          set -euo pipefail
-          npm publish --provenance --access public --tag "$TAG"
-          echo "✅ Published omniroute@$VERSION (dist-tag=$TAG) [DIRECT mode]"
+      - name: Hand the tarball to the hosted publish job
+        if: steps.resolve.outputs.skip != 'true'
+        uses: actions/upload-artifact@v7
+        with:
+          name: npm-tarball
+          path: omniroute-${{ steps.resolve.outputs.version }}.tgz
+          retention-days: 1
+          if-no-files-found: error

      - name: Publish to GitHub Packages
        if: steps.resolve.outputs.skip != 'true'
@@ -338,6 +374,117 @@ jobs:
            || echo "⚠️ omniroute@${VERSION} might already be published on GitHub Packages."
          echo "✅ Action finished for GitHub Packages"

+  # 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.
+  # v3.8.49 published fine because it predates USE_VPS_RUNNER being turned on
+  # (2026-08-02); v3.8.50 was the first release after it, so this had been latent
+  # for four weeks. Dropping --provenance was not an option: 3.8.49 carries a
+  # SLSA attestation and 3.8.50 must not regress that.
+  # The `publish` job cannot simply move to a hosted runner either — 16 GB is not
+  # enough for build:cli's next-build fallback. So it keeps proving the bytes and
+  # this job, which needs no memory at all, performs the upload.
+  stage-npm:
+    needs: publish
+    if: needs.publish.outputs.skip != 'true'
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      id-token: write # npm provenance — the whole reason this job is separate
+    steps:
+      - name: Download the tarball the publish job proved
+        uses: actions/download-artifact@v8
+        with:
+          name: npm-tarball
+          path: .
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v7
+        with:
+          node-version: ${{ env.NPM_PUBLISH_NODE_VERSION }}
+          registry-url: https://registry.npmjs.org
+
+      - name: Ensure npm supports staged publishing
+        run: |
+          set -euo pipefail
+          CUR=$(npm --version)
+          if ! node -e "const [a,b]='$(npm --version)'.split('.').map(Number); process.exit(a>11||(a===11&&b>=15)?0:1)"; then
+            # Pinned exact version (supply-chain: never float @latest in a publish
+            # job); bump deliberately when a newer npm is required.
+            echo "npm $CUR < 11.15 — installing pinned npm 11.15.0 for staged publishing"
+            npm install -g --ignore-scripts npm@11.15.0
+          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)
+        # 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 }}
+          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
+        run: |
+          set -euo pipefail
+          TARBALL="omniroute-${VERSION}.tgz"
+          test -f "$TARBALL" || { echo "tarball $TARBALL did not arrive from the publish job" >&2; ls -la; exit 1; }
+          # Always pass --tag explicitly. Defense in depth: even if VERSION is
+          # accidentally an older release, the historic tag will NOT claim `@latest`.
+          # --ignore-scripts: publishing a built tarball must never re-run
+          # prepublishOnly (build:cli-api && build:cli) on this small runner.
+          npm stage publish "$TARBALL" --provenance --access public --tag "$TAG" --ignore-scripts
+          {
+            echo "## 📦 omniroute@$VERSION STAGED (not yet installable)"
+            echo ""
+            echo "The exact bytes are parked on the registry. To release them:"
+            echo '```'
+            echo "npm stage list omniroute            # find the stage id"
+            echo "npm stage approve <id>              # owner 2FA — THE publish"
+            echo '```'
+            echo "To verify the staged bytes first: npm stage download <id> → run"
+            echo "scripts/check/check-pack-boot.mjs against them (see RELEASE_CHECKLIST)."
+            echo "To discard: npm stage reject <id>."
+          } >> "$GITHUB_STEP_SUMMARY"
+          echo "✅ Staged omniroute@$VERSION (dist-tag=$TAG) — awaiting owner 'npm stage approve'"
+
+      - name: Publish to npm (DIRECT — emergency fallback)
+        if: github.event_name == 'workflow_dispatch' && inputs.publish_mode == 'direct'
+        env:
+          VERSION: ${{ needs.publish.outputs.version }}
+          TAG: ${{ needs.publish.outputs.tag }}
+          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
+        run: |
+          set -euo pipefail
+          TARBALL="omniroute-${VERSION}.tgz"
+          test -f "$TARBALL" || { echo "tarball $TARBALL did not arrive from the publish job" >&2; ls -la; exit 1; }
+          npm publish "$TARBALL" --provenance --access public --tag "$TAG" --ignore-scripts
+          echo "✅ Published omniroute@$VERSION (dist-tag=$TAG) [DIRECT mode]"
+
  publish-opencode-plugin:
    runs-on: ubuntu-latest
    permissions:
--- a/.gitignore
+++ b/.gitignore
@@ -293,3 +293,6 @@ docker-compose.yml.bak
 # Ad-hoc test sandboxes (never tracked — may contain local DBs)
 /.sandbox/
 .aider*
+
+# check:install-upgrade work trees (~12 GB, disposable)
+/.install-upgrade/
--- a/.mailmap
+++ b/.mailmap
@@ -0,0 +1,40 @@
+# .mailmap — canonical author identities for git log/shortlog/blame.
+#
+# Why this file exists: between 2026-08-13 and 2026-08-26 this checkout carried a
+# `git config --local` whose user.name was one contributor's ("Xiangzhe" / @xz-dev)
+# and whose user.email was ANOTHER contributor's (@backryun). Every commit produced
+# on this machine in that window was therefore signed with @backryun's address —
+# 237 commits, all in the -0300 timezone, while @backryun's own work commits from
+# +0900 and continued normally throughout. The local override was removed on
+# 2026-08-26; this file repairs the RECORD without rewriting published history
+# (those commits live on release/v3.8.50 and release/v3.8.51, which other sessions
+# and open PRs build on — a rewrite would force-push both and orphan the v3.8.50 tag).
+#
+# Format: Canonical Name <canonical@email> Commit Name <commit@email>
+
+# --- Maintainer: several addresses used over the project's life ---
+diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> <diegosouza.pw@gmail.com>
+diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> <diegosouza.pw@outlook.com>
+diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> <diegosouzapw@users.noreply.github.com>
+diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> Diego Rodrigues de Sa e Souza <8016841+diegosouzapw@users.noreply.github.com>
+diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> Diego Souza <8016841+diegosouzapw@users.noreply.github.com>
+
+diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> <diego.souza.pw@gmail.com>
+diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> <souzamiriamrodrigues790@gmail.com>
+diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> <diego.souza@cdwasolutions.com.br>
+diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> <diegosouzapw@devbox.local>
+
+# --- The misattribution window: name Xiangzhe + @backryun's email, from -0300.
+# These are maintainer/session commits, NOT @backryun's contributions.
+diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> Xiangzhe <bakryun0718@proton.me>
+diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> Xiangzhe <diegosouza.pw@gmail.com>
+
+# --- Xiangzhe (@xz-dev) — a distinct contributor; keep their own work intact ---
+Xiangzhe <32761048+xz-dev@users.noreply.github.com> <xz-dev@users.noreply.github.com>
+Xiangzhe <32761048+xz-dev@users.noreply.github.com> <xiangzhedev@gmail.com>
+
+# --- @backryun's own alternate addresses (their real work, kept intact) ---
+backryun <24198422+backryun@users.noreply.github.com> <bakryun0718@proton.me>
+backryun <24198422+backryun@users.noreply.github.com> <backryun@daonlab.local>
+backryun <24198422+backryun@users.noreply.github.com> <busan011@ormbiz.co.kr>
+backryun <24198422+backryun@users.noreply.github.com> <backryun@users.noreply.github.com>
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -56,7 +56,7 @@ Repository map and Reference Documentation sections below.
 | Translators   | `open-sse/translator/`  | Format conversion (OpenAI↔Claude↔Gemini)                                                                                                                                  |
 | Transformer   | `open-sse/transformer/` | Responses API ↔ Chat Completions                                                                                                                                          |
 | Services      | `open-sse/services/`    | Combo routing, rate limits, caching, etc                                                                                                                                  |
-| Database      | `src/lib/db/`           | SQLite domain modules (159 migrations)                                                                                                                                    |
+| Database      | `src/lib/db/`           | SQLite domain modules (160 migrations)                                                                                                                                    |
 | Domain/Policy | `src/domain/`           | Policy engine, cost rules, fallback logic                                                                                                                                 |
 | MCP Server    | `open-sse/mcp-server/`  | 110 tools (44 canonical + memory/skill/GitHub/pool/gamification/plugin/Notion/Obsidian/local-corpus/RTK modules), 3 transports (stdio / SSE / Streamable HTTP), 33 scopes |
 | A2A Server    | `src/lib/a2a/`          | JSON-RPC 2.0 agent protocol                                                                                                                                               |
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -93,6 +93,67 @@

 _Living section — regenerated 2026-08-12 from all cycle commits (cycle open `ed2db6cb19` → tip). Bullets carry the merged PR and its author; direct pushes listed separately._

+### 📊 Release by the numbers
+
+| | |
+| --- | ---: |
+| 👥 People who contributed | **248** |
+| 📝 Commits in the cycle | **1,714** |
+| 🔀 Pull requests referenced | **1,666** |
+| 📋 Changelog entries | **1,182** |
+| 🙌 Contributors credited in entries | **256** |
+| 🤖 Automated dependency commits | 22 |
+
+**Entries by type**
+
+| Type | Count |
+| --- | ---: |
+| 🐛 Fixes | 779 |
+| ✨ Features | 169 |
+| 📚 Docs | 29 |
+| 🧹 Chore | 27 |
+| 🧪 Tests | 15 |
+| ♻️ Refactor | 5 |
+| ⚡ Performance | 3 |
+| providers | 2 |
+| 🔒 Security | 2 |
+| ⚙️ CI | 2 |
+| deps | 2 |
+| maint | 2 |
+
+### 🏆 Top 25 contributors this cycle
+
+_By commits in `ed2db6cb19..v3.8.50`, author identities consolidated via `.mailmap`. Bots excluded._
+
+| # | Contributor | Commits |
+| ---: | --- | ---: |
+| 🥇 | diegosouzapw | 738 |
+| 🥈 | backryun | 88 |
+| 🥉 | Dizzle | 66 |
+| 4 | Ravi Tharuma | 52 |
+| 5 | Markus Hartung | 48 |
+| 6 | Bob.Hou | 42 |
+| 7 | Rouzbeh† | 38 |
+| 8 | Xiangzhe | 31 |
+| 9 | Paco Cartones | 28 |
+| 10 | Nguyen Thanh Dat | 23 |
+| 11 | Aman | 22 |
+| 12 | Will Gordon | 19 |
+| 13 | 小妍儿 ✨ | 17 |
+| 14 | adevwithpurpose | 16 |
+| 15 | Andrew B. | 10 |
+| 16 | NOXX - Commiter | 10 |
+| 17 | ignamiranda | 10 |
+| 18 | Jonathan Bailey | 9 |
+| 19 | Ke Jin | 9 |
+| 20 | Austin Liu | 8 |
+| 21 | Chewji | 8 |
+| 22 | Prudhvi Vuda | 7 |
+| 23 | benzntech | 7 |
+| 24 | rinseaid | 7 |
+| 25 | stanley | 7 |
+
+
 ### ✨ New Features
 - **feat(search):** first-class X Search provider (`x-search`) on `POST /v1/search` and MCP `omniroute_x_search` using SuperGrok / xAI server-side `x_search`. Explicit provider or `search_type: "x"` only — never auto-selected for web. Reuses `xai-oauth` / `xao` / `xai` credentials. Not the X Developer Platform MCP. ([#10985](https://github.com/diegosouzapw/OmniRoute/issues/10985))
 - **feat(core):** add Layer A capability filter at router (#5696)
--- a/README.md
+++ b/README.md
@@ -1202,7 +1202,7 @@ Métricas canônicas em 2026-08-24: **1.029 vídeos únicos** · **11.132.922 vi
  <tr><td nowrap><b>Runtime</b></td><td>Node.js 22.x / 24.x LTS — <code>&gt;=22.22.2 &lt;23 || &gt;=24.0.0 &lt;27</code></td></tr>
  <tr><td nowrap><b>Language</b></td><td>TypeScript 6.0 — <b>100% TypeScript</b> across <code>src/</code> and <code>open-sse/</code> (zero <code>any</code> in core since v2.0)</td></tr>
  <tr><td nowrap><b>Framework</b></td><td>Next.js 16 + React 19 + Tailwind CSS 4</td></tr>
-  <tr><td nowrap><b>Database</b></td><td>better-sqlite3 (SQLite, WAL journaling) + LowDB (JSON legacy) — 120 domain modules, 159 migrations</td></tr>
+  <tr><td nowrap><b>Database</b></td><td>better-sqlite3 (SQLite, WAL journaling) + LowDB (JSON legacy) — 120 domain modules, 160 migrations</td></tr>
  <tr><td nowrap><b>Memory</b></td><td>SQLite FTS5 full-text + int8-quantized vector embeddings, typed decay</td></tr>
  <tr><td nowrap><b>Schemas</b></td><td>Zod 4 — MCP tool I/O validation + API contracts</td></tr>
  <tr><td nowrap><b>Protocols</b></td><td>MCP (stdio / HTTP / SSE) + A2A v0.3 (JSON-RPC 2.0 + SSE)</td></tr>
--- a/changelog.d/features/npm-trusted-publishing-oidc.md
+++ b/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`.
--- a/changelog.d/fixes/11741-install-upgrade-schema-convergence.md
+++ b/changelog.d/fixes/11741-install-upgrade-schema-convergence.md
@@ -0,0 +1,8 @@
+- **fix(db):** `model_capabilities` is created by a migration instead of lazily on the first
+  models.dev sync, so a clean install and an upgraded install converge on the same schema
+  regardless of which features have run
+- **fix(ci):** `check:install-upgrade` now fails on an `npm` install truncated by ENOSPC
+  (npm reports it as a warning and still exits 0), authenticates its health probe so the
+  version assertion works against the hardened health payload, frees the clean-install tree
+  before the upgrade phase, and no longer reports a schema divergence computed from a boot
+  that never served
--- a/changelog.d/fixes/11856-npm-payload-nft-manifests.md
+++ b/changelog.d/fixes/11856-npm-payload-nft-manifests.md
@@ -0,0 +1,4 @@
+- Excluded Next.js Node File Trace manifests (`*.nft.json`) from the published npm
+  tarball. They are build-time metadata and are never read while serving, but had
+  grown to 668.7 MB — 61% of the package — which pushed the upload past the
+  registry limit and made `npm publish` fail with `413 Payload Too Large`.
--- a/changelog.d/fixes/11866-alibaba-allowlist-test-timebomb.md
+++ b/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.
--- a/changelog.d/fixes/11868-npm-provenance-hosted-runner.md
+++ b/changelog.d/fixes/11868-npm-provenance-hosted-runner.md
@@ -0,0 +1,4 @@
+- Split the npm registry upload into its own GitHub-hosted job. npm refuses
+  `--provenance` from a self-hosted runner (`422 ... Only "github-hosted" runners
+  are supported`), which blocked the v3.8.50 publish; the heavy verification
+  cannot move to a hosted runner, so it now hands the proven tarball over instead.
--- a/changelog.d/fixes/v3850-electron-lockfile-build-ref.md
+++ b/changelog.d/fixes/v3850-electron-lockfile-build-ref.md
@@ -0,0 +1 @@
+- Electron release: `electron/package-lock.json` regained the optional `electron-builder-squirrel-windows` subtree (13 entries) that `npm ci` had been refusing as out of sync — the Linux desktop leg died on it — and `electron-release.yml` gained a `build_ref` dispatch input so a release whose tag was cut with the broken lock can have its assets rebuilt from the repaired line
--- a/changelog.d/fixes/v3850-electron-release-assets.md
+++ b/changelog.d/fixes/v3850-electron-release-assets.md
@@ -0,0 +1 @@
+- Electron release workflow: the `publish-npm` job now grants `actions: read` to the reusable `npm-publish.yml` it calls (its `publish` job requests it), which is what made GitHub refuse the whole v3.8.50 run at startup and ship the release with zero desktop assets; a `workflow_dispatch` now builds the requested tag instead of the dispatching branch and can skip the npm leg (`publish_npm=false`) when only re-attaching assets
--- a/changelog.d/fixes/v3850-sbom-attach-on-dispatch.md
+++ b/changelog.d/fixes/v3850-sbom-attach-on-dispatch.md
@@ -0,0 +1 @@
+- npm publish workflow: the CycloneDX SBOM is attached to the GitHub Release on `workflow_dispatch` publishes too (when a release for the tag exists), not only on the `release` event — v3.8.50 shipped through a staged dispatch and its release carried no SBOM until it was attached by hand from the run's `sbom-npm` artifact
--- a/changelog.d/maintenance/11866-config-expiry-time-bomb-test.md
+++ b/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.
--- a/changelog.d/maintenance/11878-check-workflows-provenance-self-hosted.md
+++ b/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.
--- a/changelog.d/maintenance/11892-runner-janitor-act-not-advise.md
+++ b/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).
--- a/changelog.d/maintenance/11896-ci-artifact-download-to-disk.md
+++ b/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.
--- a/changelog.d/maintenance/11897-ci-heavy-build-lane.md
+++ b/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.
--- a/changelog.d/maintenance/ci-coverage-codecov-step-timeout.md
+++ b/changelog.d/maintenance/ci-coverage-codecov-step-timeout.md
@@ -0,0 +1 @@
+- `Coverage` job on `ci.yml`: the informational Codecov upload gets its own 5-minute ceiling and `continue-on-error`, and the job budget grows from 20 to 30 minutes (the 8-shard c8 merge alone takes ~10) — a stalled upload no longer ends the job `cancelled` and drags a fully green `main` run's conclusion down with it
--- a/changelog.d/maintenance/ci-omni-build-runner-label.md
+++ b/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.
--- a/config/quality/install-upgrade-allowlist.json
+++ b/config/quality/install-upgrade-allowlist.json
@@ -1,6 +1,6 @@
 {
-  "_doc": "Tables that exist ONLY in databases upgraded from an older version — residue whose CREATE left the migration set in some past cycle but survives where it already existed. Harmless (nothing references them), but recorded here so check-install-upgrade.mjs can still fail on a NEW divergence. The opposite direction (a table a clean install creates but an upgrade does not) is NEVER allowlisted: it means every existing user is missing structure the code expects.",
+  "_doc": "Tables that exist ONLY in databases upgraded from an older version. Two causes, and they call for different fixes: (a) residue whose CREATE left the migration set in some past cycle but survives where it already existed — allowlist it here; (b) a table created LAZILY at runtime with `CREATE TABLE IF NOT EXISTS` inside a feature code path — whether a database has it depends on whether that feature ran, so it diverges by TIMING and can show up on EITHER side. Fix (b) with a migration instead of an entry here (see src/lib/db/migrations/163_model_capabilities.sql); an allowlist entry only hides it in one direction. Either way, recorded so check-install-upgrade.mjs can still fail on a NEW divergence. The opposite direction (a table a clean install creates but an upgrade does not) is NEVER allowlisted: it means every existing user is missing structure the code expects.",
  "residualTables": {
-    "cache_metrics": "Measured 2026-07-30 on a real 3.8.48 install upgraded to 3.8.49 (VPS .16, 165 MB database, 114 → 117 tables). Present in upgraded databases, absent from clean installs. No code path referenced it during the upgrade (zero `no such table` in 150 log lines, both installs healthy). Left in place rather than dropped: a DROP migration on a table we cannot prove is unused everywhere is the riskier change. Revisit when the cache subsystem is next touched."
+    "cache_metrics": "Measured 2026-07-30 on a real 3.8.48 install upgraded to 3.8.49 (VPS .16, 165 MB database, 114 → 117 tables). Present in upgraded databases, absent from clean installs. Cause identified 2026-08-27: it is case (b) above — created lazily by `ensureCacheMetricsTable()` at src/lib/semanticCache.ts:34, never by a migration, so it appears only where the semantic cache has run. No code path referenced it during the upgrade (zero `no such table` in 150 log lines, both installs healthy). Left as an allowlist entry rather than promoted to a migration or dropped: unlike model_capabilities it did not block a release, and creating a table for a subsystem we cannot prove is live is not a change to make blind. Revisit when the cache subsystem is next touched."
  }
 }
--- a/config/quality/quality-baseline.json
+++ b/config/quality/quality-baseline.json
@@ -169,7 +169,7 @@
      "dedicatedGate": true
    },
    "zizmorFindings": {
-      "value": 192,
+      "value": 194,
      "_rebaseline_2026_08_20_radar_export_workflow": "190 -> 192 (+2). Workflow novo `.github/workflows/radar-export.yml` (passo 10 do go-live do Radar: publica o export estável do catálogo como asset de release para o servidor privado baixar via RADAR_EXPORT_URL). Os +2 são unpinned-uses @vN: actions/checkout@v7 + actions/setup-node@v7 — a MESMA convenção deliberada de todos os workflows (ver _scanner_harden_workflows_2026_06_16); fixar por SHA só este violaria a convenção. O findings artipacked do checkout foi CORRIGIDO com `persist-credentials: false` (o job publica via GH_TOKEN em `gh release`, não usa a credencial do checkout). Nenhuma classe nova de template-injection / cache-poisoning / dangerous-triggers. Medido local com zizmor 1.25.2 via `node scripts/check/check-workflows.mjs --ratchet` = 191; +1 do delta conhecido do runner (ver _rebaseline_2026_07_28_ci_runner_delta: o runner enxerga 1 unpinned-uses @vN a mais que o devbox no mesmo commit; a baseline segue o runner) => 192.",
      "_rebaseline_2026_07_20_aliasresolver_hook_split_7808": "175 -> 176 (+1). Companion to PR #7808 (CodeQL js/incomplete-url-substring-sanitization fix in bin/aliasResolver.mjs). The +1 is NOT caused by this PR's code changes (bin/* is not a workflow file) — it is a pre-existing drift that surfaced because the ratchet gate runs on this PR's CI: the zizmor scanner version on the GitHub runner gained a new rule (or extended an existing one) since the v3.8.49 baseline was seeded on 2026-07-17. Breakdown: the new finding is an unpinned-uses @vN class item on one of the existing workflows (same deliberate convention as _scanner_harden_workflows_2026_06_16 — @vN is intentional, SHA-pinning only this one would violate the convention). No new template-injection/artipacked/cache-poisoning/dangerous-triggers classes introduced. Measured by the Quality Gates (Extended) job on run 29713001401 = 176, baseline was 175. Note: by the time this landed on release/v3.8.49, the baseline was already at 176 via _rebaseline_2026_07_17_combo_recovery_hints — this entry is kept as historical record; no further bump applied.",
      "_rebaseline_2026_07_17_v3849_release": "169 -> 175 (+6). Cycle workflow drift (v3.8.48/v3.8.49): npm-publish.yml (new, WS1.3 #7092), electron-release.yml, nightly-compat.yml, nightly-release-green.yml, CI restructures (#7501 full-history base fetch, #7355 main-green, #7202 merge-queue gates, Trunk/Codecov). Breakdown vs v3.8.47: +3 unpinned-uses (@vN convention, deliberate per _scanner_harden_workflows_2026_06_16), +2 cache-poisoning (artifact upload/cache in the OWN electron-release/npm-publish RELEASE workflows -- operator-controlled, not fork-PR exploitable), +1 excessive-permissions (nightly-compat.yml permissions:issues). No new template-injection/artipacked/dangerous-triggers. Measured with zizmor 1.25.2 via `node scripts/check/check-workflows.mjs --ratchet` = 175 on da3a0be69.",
@@ -180,7 +180,8 @@
      "_rebaseline_2026_06_23_v3834_release": "152 -> 155 (+3). The 3 new unpinned-uses are in .github/workflows/nightly-release-green.yml (added by #4622 this cycle): actions/checkout@v7, actions/setup-node@v6, actions/upload-artifact@v4 — the SAME deliberate @vN convention as ci.yml's own checkout@v7/setup-node@v6 and every other workflow (see _scanner_harden_workflows_2026_06_16 + _zizmor_rebaseline_2026_06_20_ci_build_artifact_reuse). SHA-pinning only this workflow would violate the convention. The workflow-lint ratchet does NOT run on PR->release fast-gates, so it surfaced only on the release PR; measured locally via `npm run check:workflows -- --ratchet` = 155. No new template-injection/artipacked/cache-poisoning.",
      "_rebaseline_2026_07_13_v3847_release_preflight": "159 -> 169 (+10). Findings from cycle-merged workflow changes: #6716 (PR gate restructure), #6781 (unit fast-path shard 2->4), #6788 (TIA tsx loader split), #6881 (electron-updater latest.yml manifests in release assets) — same deliberate @vN unpinned-uses convention as prior rebaselines; no new template-injection/artipacked/cache-poisoning classes. Measured via `npm run check:workflows -- --ratchet` = 169 on the v3.8.47 release pre-flight.",
      "_rebaseline_2026_07_28_v3849_release_preflight": "176 -> 189 (+13). Pre-flight de fechamento da v3.8.49 (934 commits no ciclo). Deriva de workflow: 1 workflow novo (build-rinseaid-image.yml) mais os bumps de action do Dependabot ao longo do ciclo — todos da MESMA classe unpinned-uses @vN, convenção deliberada do repo (ver _scanner_harden_workflows_2026_06_16); fixar por SHA só estes violaria a convenção. Nenhuma classe nova de template-injection / artipacked / cache-poisoning / dangerous-triggers. Nesta mesma passada foram CORRIGIDAS 3 diretivas shellcheck malformadas (SC1125: `# shellcheck disable=SC2086 — texto`, em que o travessão invalida o par key=value) em ci.yml e nightly-release-green.yml. Medido com zizmor 1.25.2 via `npm run check:workflows -- --ratchet` = 189.",
-      "_rebaseline_2026_07_28_ci_runner_delta": "189 -> 190 (+1). Medido 189 no devbox e 190 no runner do GitHub no MESMO commit (run 30396592013, job Quality Gates (Extended)) — mesma classe já registrada em _rebaseline_2026_07_20_aliasresolver_hook_split_7808: a versão do zizmor no runner enxerga uma finding a mais que a local, sempre da classe unpinned-uses @vN. O valor do runner é o que o gate compara, então a baseline segue o runner."
+      "_rebaseline_2026_07_28_ci_runner_delta": "189 -> 190 (+1). Medido 189 no devbox e 190 no runner do GitHub no MESMO commit (run 30396592013, job Quality Gates (Extended)) — mesma classe já registrada em _rebaseline_2026_07_20_aliasresolver_hook_split_7808: a versão do zizmor no runner enxerga uma finding a mais que a local, sempre da classe unpinned-uses @vN. O valor do runner é o que o gate compara, então a baseline segue o runner.",
+      "_rebaseline_2026_08_28_npm_publish_hosted_stage_job": "192 -> 194 (+2). Job novo `stage-npm` em .github/workflows/npm-publish.yml: o npm RECUSA `--provenance` vindo de runner self-hosted (422 \"Unsupported GitHub Actions runner environment\"), e o job `publish` nao pode migrar para runner hospedado porque 16 GB nao bastam para o fallback next-build do build:cli (documentado no proprio runs-on). A separacao foi a unica saida que preserva a atestacao SLSA que a 3.8.49 ja tem. Os +2 sao da MESMA convencao deliberada de todos os workflows (ver _scanner_harden_workflows_2026_06_16): unpinned-uses @vN em actions/download-artifact@v8 + actions/setup-node@v7, mais o cache-poisoning que o proprio setup-node@v7 ja gera nos outros 2 jobs deste MESMO arquivo (linhas 85 e 463) e que ja esta na baseline. Fixar por SHA so este job violaria a convencao. Nenhuma classe nova: zero template-injection / artipacked / dangerous-triggers / excessive-permissions — o job declara apenas contents:read + id-token:write, que e o minimo para a proveniencia. Medido pelo job Quality Gates (Extended) no run 33162... da PR #11877 = 194."
    },
    "vulnCount": {
      "value": 22,
--- a/docs/i18n/ar/llm.txt
+++ b/docs/i18n/ar/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/az/llm.txt
+++ b/docs/i18n/az/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/bg/llm.txt
+++ b/docs/i18n/bg/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/bn/llm.txt
+++ b/docs/i18n/bn/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/cs/llm.txt
+++ b/docs/i18n/cs/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/da/llm.txt
+++ b/docs/i18n/da/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/de/llm.txt
+++ b/docs/i18n/de/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/es/llm.txt
+++ b/docs/i18n/es/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/fa/llm.txt
+++ b/docs/i18n/fa/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/fi/llm.txt
+++ b/docs/i18n/fi/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/fr/llm.txt
+++ b/docs/i18n/fr/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/gu/llm.txt
+++ b/docs/i18n/gu/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/he/llm.txt
+++ b/docs/i18n/he/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/hi/llm.txt
+++ b/docs/i18n/hi/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/hu/llm.txt
+++ b/docs/i18n/hu/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/id/llm.txt
+++ b/docs/i18n/id/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/in/llm.txt
+++ b/docs/i18n/in/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/it/llm.txt
+++ b/docs/i18n/it/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/ja/llm.txt
+++ b/docs/i18n/ja/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/ko/llm.txt
+++ b/docs/i18n/ko/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/mr/llm.txt
+++ b/docs/i18n/mr/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/ms/llm.txt
+++ b/docs/i18n/ms/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/nl/llm.txt
+++ b/docs/i18n/nl/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/no/llm.txt
+++ b/docs/i18n/no/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/phi/llm.txt
+++ b/docs/i18n/phi/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/pl/llm.txt
+++ b/docs/i18n/pl/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/pt-BR/llm.txt
+++ b/docs/i18n/pt-BR/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/pt/llm.txt
+++ b/docs/i18n/pt/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/ro/llm.txt
+++ b/docs/i18n/ro/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/ru/llm.txt
+++ b/docs/i18n/ru/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/sk/llm.txt
+++ b/docs/i18n/sk/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/sv/llm.txt
+++ b/docs/i18n/sv/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/sw/llm.txt
+++ b/docs/i18n/sw/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/ta/llm.txt
+++ b/docs/i18n/ta/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/te/llm.txt
+++ b/docs/i18n/te/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/th/llm.txt
+++ b/docs/i18n/th/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/tr/llm.txt
+++ b/docs/i18n/tr/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/uk-UA/llm.txt
+++ b/docs/i18n/uk-UA/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/ur/llm.txt
+++ b/docs/i18n/ur/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/vi/llm.txt
+++ b/docs/i18n/vi/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/zh-CN/llm.txt
+++ b/docs/i18n/zh-CN/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/i18n/zh-TW/llm.txt
+++ b/docs/i18n/zh-TW/llm.txt
@@ -18,7 +18,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -438,7 +438,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/docs/ops/RELEASE_CHECKLIST.md
+++ b/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
--- a/docs/ops/RUNNER_BOX.md
+++ b/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.
--- a/docs/reference/ENVIRONMENT.md
+++ b/docs/reference/ENVIRONMENT.md
@@ -103,6 +103,7 @@ OmniRoute uses **SQLite** (via `better-sqlite3`) for all persistence. These vari
 | `OMNIROUTE_MIGRATIONS_DIR`             | _(auto-detect)_      | `src/lib/db/migrationRunner.ts`                       | Override the directory that the migration runner scans. Useful when shipping bundled migrations in custom builds.                                                                                                                               |
 | `OMNIROUTE_EXTRA_MIGRATIONS_DIRS`      | _(unset)_            | `src/lib/db/migrationRunner/extraDirs.ts`             | Additional migration directories as `namespace=dir` entries separated by the platform path delimiter (e.g. `ee=/opt/app/enterprise/db/migrations`). Files found there are recorded as `<namespace>-<number>`, so a distribution shipping its own migrations never collides with the upstream numeric slots. A malformed entry, an invalid namespace or a missing directory throws at startup instead of silently skipping the schema. |
 | `OMNIROUTE_MAX_PENDING_MIGRATIONS`     | `50`                 | `src/lib/db/migrationRunner.ts`                       | Mass-pending-migrations safety threshold (#3416). Startup aborts if more than this many migrations are pending on an existing DB (guards against a wiped tracking table). Raise it to restore an older backup; set to `0` to disable the check. |
+| `OMNIROUTE_INSTALL_UPGRADE_WORKDIR`     | _(`<repo>/.install-upgrade`)_ | `scripts/check/check-install-upgrade.mjs`             | Working directory for the `check:install-upgrade` release gate. It needs roughly 12 GB (two ~3 GB install trees plus the tarball), so it must not run on a small tmpfs — on the self-hosted runner `/tmp` is a 12 GB RAM-backed tmpfs and the gate exhausted it, truncating the package. |
 | `OMNIROUTE_SPEND_FLUSH_INTERVAL_MS`    | _(default in code)_  | `src/lib/spend/batchWriter.ts`                        | Flush interval (ms) for the batched spend/cost writer. Lower values reduce write coalescing; higher values reduce DB contention.                                                                                                                |
 | `OMNIROUTE_SPEND_MAX_BUFFER_SIZE`      | _(default in code)_  | `src/lib/spend/batchWriter.ts`                        | Max buffered spend entries before a forced flush. Raise on high-QPS deployments; lower when bounded memory matters more.                                                                                                                        |
 | `OMNIROUTE_PROXY_FETCH_DEBUG`          | _(unset)_            | `open-sse/utils/proxyFetch.ts`                        | Set to `"true"` to emit `[ProxyFetch]` debug logs on the Vercel relay path. Off by default to avoid leaking routing hints.                                                                                                                      |
--- a/electron/package-lock.json
+++ b/electron/package-lock.json
@@ -297,6 +297,45 @@
        "url": "https://github.com/sponsors/isaacs"
      }
    },
+    "node_modules/@electron/windows-sign": {
+      "version": "1.2.2",
+      "resolved": "https://registry.npmjs.org/@electron/windows-sign/-/windows-sign-1.2.2.tgz",
+      "integrity": "sha512-dfZeox66AvdPtb2lD8OsIIQh12Tp0GNCRUDfBHIKGpbmopZto2/A8nSpYYLoedPIHpqkeblZ/k8OV0Gy7PYuyQ==",
+      "dev": true,
+      "license": "BSD-2-Clause",
+      "optional": true,
+      "peer": true,
+      "dependencies": {
+        "cross-dirname": "^0.1.0",
+        "debug": "^4.3.4",
+        "fs-extra": "^11.1.1",
+        "minimist": "^1.2.8",
+        "postject": "^1.0.0-alpha.6"
+      },
+      "bin": {
+        "electron-windows-sign": "bin/electron-windows-sign.js"
+      },
+      "engines": {
+        "node": ">=14.14"
+      }
+    },
+    "node_modules/@electron/windows-sign/node_modules/fs-extra": {
+      "version": "11.4.0",
+      "resolved": "https://registry.npmjs.org/fs-extra/-/fs-extra-11.4.0.tgz",
+      "integrity": "sha512-EQsFzMUJkCKGr1ePqlYADkIUmHW1s3ZXr5Yqy6wbGrfUCphpl2maM/kyOIRA2HpP3AaFQTZXD4ldjek+nccddA==",
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "peer": true,
+      "dependencies": {
+        "graceful-fs": "^4.2.0",
+        "jsonfile": "^6.0.1",
+        "universalify": "^2.0.0"
+      },
+      "engines": {
+        "node": ">=14.14"
+      }
+    },
    "node_modules/@isaacs/fs-minipass": {
      "version": "4.0.1",
      "resolved": "https://registry.npmjs.org/@isaacs/fs-minipass/-/fs-minipass-4.0.1.tgz",
@@ -1091,6 +1130,15 @@
      "dev": true,
      "license": "MIT"
    },
+    "node_modules/cross-dirname": {
+      "version": "0.1.0",
+      "resolved": "https://registry.npmjs.org/cross-dirname/-/cross-dirname-0.1.0.tgz",
+      "integrity": "sha512-+R08/oI0nl3vfPcqftZRpytksBXDzOUveBq/NBVx0sUp1axwzPQrKinNx5yd5sxPu8j1wIy8AfnVQ+5eFdha6Q==",
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "peer": true
+    },
    "node_modules/cross-spawn": {
      "version": "7.0.6",
      "resolved": "https://registry.npmjs.org/cross-spawn/-/cross-spawn-7.0.6.tgz",
@@ -1411,6 +1459,19 @@
        "node": ">=14.0.0"
      }
    },
+    "node_modules/electron-builder-squirrel-windows": {
+      "version": "26.15.3",
+      "resolved": "https://registry.npmjs.org/electron-builder-squirrel-windows/-/electron-builder-squirrel-windows-26.15.3.tgz",
+      "integrity": "sha512-Jc19XPV9y9+2bAdZPkXuVNGNIEFBq9poHC61l8Kv6FdK7DRG3+Ic0rerC0DXOaeHNz8yW0fg/JnF8GQROOF5MA==",
+      "dev": true,
+      "license": "MIT",
+      "peer": true,
+      "dependencies": {
+        "app-builder-lib": "26.15.3",
+        "builder-util": "26.15.3",
+        "electron-winstaller": "5.4.0"
+      }
+    },
    "node_modules/electron-publish": {
      "version": "26.15.3",
      "resolved": "https://registry.npmjs.org/electron-publish/-/electron-publish-26.15.3.tgz",
@@ -1445,6 +1506,66 @@
        "tiny-typed-emitter": "^2.1.0"
      }
    },
+    "node_modules/electron-winstaller": {
+      "version": "5.4.0",
+      "resolved": "https://registry.npmjs.org/electron-winstaller/-/electron-winstaller-5.4.0.tgz",
+      "integrity": "sha512-bO3y10YikuUwUuDUQRM4KfwNkKhnpVO7IPdbsrejwN9/AABJzzTQ4GeHwyzNSrVO+tEH3/Np255a3sVZpZDjvg==",
+      "dev": true,
+      "hasInstallScript": true,
+      "license": "MIT",
+      "peer": true,
+      "dependencies": {
+        "@electron/asar": "^3.2.1",
+        "debug": "^4.1.1",
+        "fs-extra": "^7.0.1",
+        "lodash": "^4.17.21",
+        "temp": "^0.9.0"
+      },
+      "engines": {
+        "node": ">=8.0.0"
+      },
+      "optionalDependencies": {
+        "@electron/windows-sign": "^1.1.2"
+      }
+    },
+    "node_modules/electron-winstaller/node_modules/fs-extra": {
+      "version": "7.0.1",
+      "resolved": "https://registry.npmjs.org/fs-extra/-/fs-extra-7.0.1.tgz",
+      "integrity": "sha512-YJDaCJZEnBmcbw13fvdAM9AwNOJwOzrE4pqMqBq5nFiEqXUqHwlK4B+3pUw6JNvfSPtX05xFHtYy/1ni01eGCw==",
+      "dev": true,
+      "license": "MIT",
+      "peer": true,
+      "dependencies": {
+        "graceful-fs": "^4.1.2",
+        "jsonfile": "^4.0.0",
+        "universalify": "^0.1.0"
+      },
+      "engines": {
+        "node": ">=6 <7 || >=8"
+      }
+    },
+    "node_modules/electron-winstaller/node_modules/jsonfile": {
+      "version": "4.0.0",
+      "resolved": "https://registry.npmjs.org/jsonfile/-/jsonfile-4.0.0.tgz",
+      "integrity": "sha512-m6F1R3z8jjlf2imQHS2Qez5sjKWQzbuuhuJ/FKYFRZvPE3PuHcSMVZzfsLhGVOkfd20obL5SWEBew5ShlquNxg==",
+      "dev": true,
+      "license": "MIT",
+      "peer": true,
+      "optionalDependencies": {
+        "graceful-fs": "^4.1.6"
+      }
+    },
+    "node_modules/electron-winstaller/node_modules/universalify": {
+      "version": "0.1.2",
+      "resolved": "https://registry.npmjs.org/universalify/-/universalify-0.1.2.tgz",
+      "integrity": "sha512-rBJeI5CXAlmy1pV+617WB9J63U6XcazHHF2f2dbJix4XzpUF0RS3Zbj0FGIOCAva5P/d/GBOYaACQ1w+0azUkg==",
+      "dev": true,
+      "license": "MIT",
+      "peer": true,
+      "engines": {
+        "node": ">= 4.0.0"
+      }
+    },
    "node_modules/emoji-regex": {
      "version": "8.0.0",
      "resolved": "https://registry.npmjs.org/emoji-regex/-/emoji-regex-8.0.0.tgz",
@@ -2359,6 +2480,20 @@
        "node": ">= 18"
      }
    },
+    "node_modules/mkdirp": {
+      "version": "0.5.6",
+      "resolved": "https://registry.npmjs.org/mkdirp/-/mkdirp-0.5.6.tgz",
+      "integrity": "sha512-FP+p8RB8OWpF3YZBCrP5gtADmtXApB5AMLn+vdyA+PyxCjrCs00mjyUozssO33cwDeT3wNGdLxJ5M//YqtHAJw==",
+      "dev": true,
+      "license": "MIT",
+      "peer": true,
+      "dependencies": {
+        "minimist": "^1.2.6"
+      },
+      "bin": {
+        "mkdirp": "bin/cmd.js"
+      }
+    },
    "node_modules/ms": {
      "version": "2.1.3",
      "resolved": "https://registry.npmjs.org/ms/-/ms-2.1.3.tgz",
@@ -2622,6 +2757,36 @@
        "node": ">=18"
      }
    },
+    "node_modules/postject": {
+      "version": "1.0.0-alpha.6",
+      "resolved": "https://registry.npmjs.org/postject/-/postject-1.0.0-alpha.6.tgz",
+      "integrity": "sha512-b9Eb8h2eVqNE8edvKdwqkrY6O7kAwmI8kcnBv1NScolYJbo59XUF0noFq+lxbC1yN20bmC0WBEbDC5H/7ASb0A==",
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "peer": true,
+      "dependencies": {
+        "commander": "^9.4.0"
+      },
+      "bin": {
+        "postject": "dist/cli.js"
+      },
+      "engines": {
+        "node": ">=14.0.0"
+      }
+    },
+    "node_modules/postject/node_modules/commander": {
+      "version": "9.5.0",
+      "resolved": "https://registry.npmjs.org/commander/-/commander-9.5.0.tgz",
+      "integrity": "sha512-KRs7WVDKg86PWiuAqhDrAQnTXZKraVcCc6vFdL14qrZ/DcWwuRo7VoiYXalXO7S5GKpqYiVEwCbgFDfxNHKJBQ==",
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "peer": true,
+      "engines": {
+        "node": "^12.20.0 || >=14"
+      }
+    },
    "node_modules/proc-log": {
      "version": "6.1.0",
      "resolved": "https://registry.npmjs.org/proc-log/-/proc-log-6.1.0.tgz",
@@ -2816,6 +2981,21 @@
        "node": ">= 4"
      }
    },
+    "node_modules/rimraf": {
+      "version": "2.6.3",
+      "resolved": "https://registry.npmjs.org/rimraf/-/rimraf-2.6.3.tgz",
+      "integrity": "sha512-mwqeW5XsA2qAejG46gYdENaxXjx9onRNCfn7L0duuP4hCuTIi/QO7PDK07KJfp1d+izWPrzEJDcSqBa0OZQriA==",
+      "deprecated": "Rimraf versions prior to v4 are no longer supported",
+      "dev": true,
+      "license": "ISC",
+      "peer": true,
+      "dependencies": {
+        "glob": "^7.1.3"
+      },
+      "bin": {
+        "rimraf": "bin.js"
+      }
+    },
    "node_modules/roarr": {
      "version": "2.15.4",
      "resolved": "https://registry.npmjs.org/roarr/-/roarr-2.15.4.tgz",
@@ -3071,6 +3251,21 @@
        "node": ">=18"
      }
    },
+    "node_modules/temp": {
+      "version": "0.9.4",
+      "resolved": "https://registry.npmjs.org/temp/-/temp-0.9.4.tgz",
+      "integrity": "sha512-yYrrsWnrXMcdsnu/7YMYAofM1ktpL5By7vZhf15CrXijWWrEYZks5AXBudalfSWJLlnen/QUJUB5aoB0kqZUGA==",
+      "dev": true,
+      "license": "MIT",
+      "peer": true,
+      "dependencies": {
+        "mkdirp": "^0.5.1",
+        "rimraf": "~2.6.2"
+      },
+      "engines": {
+        "node": ">=6.0.0"
+      }
+    },
    "node_modules/temp-file": {
      "version": "3.4.0",
      "resolved": "https://registry.npmjs.org/temp-file/-/temp-file-3.4.0.tgz",
--- a/llm.txt
+++ b/llm.txt
@@ -14,7 +14,7 @@ OmniRoute solves the problem of managing multiple AI provider subscriptions, quo

 - **Runtime:** Node.js `>=22.0.0 <23 || >=24.0.0 <27`, ES Modules (`"type": "module"`)
 - **Framework:** Next.js 16 (App Router) with TypeScript 6
- **Database:** SQLite via better-sqlite3 (local, zero-config, 159 migrations)
+- **Database:** SQLite via better-sqlite3 (local, zero-config, 160 migrations)
 - **State management:** Zustand (client), SQLite (server persistence)
 - **UI:** React 19, Tailwind CSS 4, Recharts for analytics, @lobehub/icons for 130+ provider SVG icons
 - **Auth:** OAuth 2.0 (PKCE) for providers, bcrypt for local user auth
@@ -434,7 +434,7 @@ diagnostics) plus **memory**, **skill**, **agentSkill**, **githubSkill**, **pool

 4. **Environment variables:** All configuration is in `.env` (from `.env.example`). Key vars: `PORT`, `NEXT_PUBLIC_BASE_URL`, `API_KEY`, `ADMIN_PASSWORD`.

-5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 159 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.
+5. **Database layer:** Operations go through `src/lib/db/` modules (120 domain-specific files, 160 migrations). `localDb.ts` is re-exports only — add new functions to the proper `db/*.ts` module.

 6. **Tests** use Node.js built-in test runner + Vitest. Run `npm test`. Vitest for MCP/autoCombo (`npm run test:vitest`). Playwright for E2E (`npm run test:e2e`). Coverage gate: ratchet vs `quality-baseline.json`, absolute floor 60% statements/lines/functions/branches.

--- a/package.json
+++ b/package.json
@@ -52,7 +52,8 @@
    "!**/*.test.js",
    "!**/*.test.mjs",
    "!**/*.spec.ts",
-    "!**/*.spec.tsx"
+    "!**/*.spec.tsx",
+    "!**/*.nft.json"
  ],
  "workspaces": [
    "open-sse",
--- a/scripts/check/check-install-upgrade.mjs
+++ b/scripts/check/check-install-upgrade.mjs
@@ -30,10 +30,12 @@
 * Requires `npm run build:cli` first — this is a --with-build gate, like check:pack-boot.
 */

-import { execFileSync, spawn } from "node:child_process";
+import { execFileSync, spawn, spawnSync } from "node:child_process";
+import crypto from "node:crypto";
 import fs from "node:fs";
 import os from "node:os";
 import path from "node:path";
+import { pathToFileURL } from "node:url";
 import { DatabaseSync } from "node:sqlite";

 const BOOT_DEADLINE_MS = 180_000;
@@ -43,6 +45,59 @@ const ALLOWLIST_PATH = "config/quality/install-upgrade-allowlist.json";
 const log = (msg) => console.log(`[install-upgrade] ${msg}`);
 const warn = (msg) => console.log(`[install-upgrade] ⚠️  ${msg}`);

+/** Root of the installed package inside an `npm install -g --prefix` tree. */
+function packageRootFor(prefix) {
+  return path.join(prefix, "lib", "node_modules", "omniroute");
+}
+
+/**
+ * Credential for the health probe.
+ *
+ * GHSA-mvf8-qc78-5mxm hardened /api/monitoring/health: an ANONYMOUS caller now gets only
+ * `{ status }` — the version, node version, pid and provider config are reserved for a
+ * management principal (src/app/api/monitoring/health/route.ts → publicHealthView). An
+ * unauthenticated probe therefore reads `body.version === undefined`, and this gate's
+ * version assertion could never pass again; the v3.8.50 publish run failed with
+ * "clean: health reports version undefined, expected 3.8.50" for exactly that reason.
+ *
+ * The gate spawns the server itself, so it can mint the credential instead of guessing one:
+ * `OMNIROUTE_INTERNAL_SERVICE_TOKEN` + the `x-omniroute-internal-service-token` header is
+ * accepted by requireManagementAuth() via isTrustedLoopbackInternalServiceRequest(), and the
+ * probe is loopback by construction. Unlike the machine token `check:pack-boot` derives, this
+ * does not depend on a readable machine-id, and an older PREVIOUS version that never gated
+ * health simply ignores the header. The assertion keeps its full strength — it just presents
+ * a credential.
+ */
+const INTERNAL_SERVICE_TOKEN = crypto.randomBytes(32).toString("hex");
+const INTERNAL_SERVICE_HEADER = "x-omniroute-internal-service-token";
+
+/**
+ * Secondary credential: the same loopback machine token `check:pack-boot` derives from the
+ * packaged CLI. Sent alongside the internal-service token so a build that only honours one
+ * of the two still answers with the full payload.
+ */
+function derivePackagedCliToken(prefix) {
+  const cliModuleUrl = pathToFileURL(
+    path.join(packageRootFor(prefix), "bin", "cli", "utils", "cliToken.mjs")
+  ).href;
+  try {
+    return execFileSync(
+      process.execPath,
+      [
+        "--input-type=module",
+        "--eval",
+        "import(process.argv[1]).then(async m => process.stdout.write(await m.getCliToken()))",
+        cliModuleUrl,
+      ],
+      { encoding: "utf8", env: { ...process.env } }
+    ).trim();
+  } catch {
+    // A truncated/broken install cannot derive a token. Returning null keeps the boot
+    // probe running (it will fail loudly on its own) instead of crashing the gate here.
+    return null;
+  }
+}
+
 function pickTarball(packJson) {
  const filename = JSON.parse(packJson)?.[0]?.filename;
  if (!filename) throw new Error("npm pack --json returned no filename");
@@ -121,6 +176,11 @@ async function bootAndProbe({ prefix, dataDir, port, expectVersion, label }) {
  if (!fs.existsSync(binPath)) {
    return { ok: false, failures: [`${label}: bin not found at ${binPath}`], tail: [] };
  }
+  const cliToken = derivePackagedCliToken(prefix);
+  const probeHeaders = {
+    [INTERNAL_SERVICE_HEADER]: INTERNAL_SERVICE_TOKEN,
+    ...(cliToken ? { "x-omniroute-cli-token": cliToken } : {}),
+  };
  const child = spawn(binPath, ["serve", "--port", String(port)], {
    env: {
      ...process.env,
@@ -130,6 +190,7 @@ async function bootAndProbe({ prefix, dataDir, port, expectVersion, label }) {
      API_KEY_SECRET: "install-upgrade-gate-api-key-secret-long",
      DISABLE_SQLITE_AUTO_BACKUP: "true",
      OMNIROUTE_SKIP_SYSTEM_TRUST: "1",
+      OMNIROUTE_INTERNAL_SERVICE_TOKEN: INTERNAL_SERVICE_TOKEN,
    },
    stdio: ["ignore", "pipe", "pipe"],
    detached: true,
@@ -151,20 +212,33 @@ async function bootAndProbe({ prefix, dataDir, port, expectVersion, label }) {
  let result = { ok: false, failures: [`${label}: never became healthy`], tail };
  while (Date.now() < deadline) {
    if (childExit !== null) {
-      result = { ok: false, failures: [`${label}: exited with code ${childExit} before serving`], tail };
+      result = {
+        ok: false,
+        failures: [`${label}: exited with code ${childExit} before serving`],
+        tail,
+      };
      break;
    }
    try {
-      const res = await fetch(`http://127.0.0.1:${port}/api/monitoring/health`);
+      const res = await fetch(`http://127.0.0.1:${port}/api/monitoring/health`, {
+        headers: probeHeaders,
+      });
      const body = await res.json().catch(() => null);
      if (res.status === 200 && body && typeof body === "object") {
        const failures = [];
        // `status` may legitimately report degraded (no providers configured) — the gate
        // targets boot crashes and version mismatches, not health of a bare install.
-        if (expectVersion && body.version !== expectVersion) {
-          failures.push(`${label}: health reports version ${body.version}, expected ${expectVersion}`);
+        const reportedVersion = body.version ?? body.system?.version;
+        if (expectVersion && reportedVersion !== expectVersion) {
+          failures.push(
+            `${label}: health reports version ${reportedVersion}, expected ${expectVersion}` +
+              (reportedVersion === undefined
+                ? " — the payload carries no version at all, which is the ANONYMOUS health " +
+                  "view: the probe's credentials were not accepted (see GHSA-mvf8-qc78-5mxm)"
+                : "")
+          );
        }
-        result = { ok: failures.length === 0, version: body.version, failures, tail };
+        result = { ok: failures.length === 0, version: reportedVersion, failures, tail };
        break;
      }
    } catch {
@@ -183,13 +257,72 @@ async function bootAndProbe({ prefix, dataDir, port, expectVersion, label }) {
  return result;
 }

-function npmInstallInto(prefix, spec) {
-  execFileSync("npm", ["install", "-g", "--prefix", prefix, "--no-audit", "--no-fund", spec], {
-    encoding: "utf8",
-    maxBuffer: 128 * 1024 * 1024,
-  });
+/**
+ * `npm install` reports ENOSPC as a *warning* per failed tar entry and still exits 0.
+ *
+ * That is not a theoretical concern: on the v3.8.50 publish run the Phase B upgrade install
+ * emitted 5611 `npm warn tar TAR_ENTRY_ERROR ENOSPC: no space left on device` lines, exited
+ * 0, and left a truncated package behind. `omniroute serve` then "exited with code 0 before
+ * serving", no migration ever ran, and the gate concluded the release was missing 15 tables
+ * — a full false alarm produced by a full disk. Each install tree is ~3 GB, and the run
+ * builds two of them plus a ~275 MB tarball.
+ *
+ * So: surface the truncation at the install, where it is unambiguous.
+ */
+function npmInstallInto(prefix, spec, label = spec) {
+  // spawnSync (not execFileSync): execFileSync forwards the child's stderr straight to the
+  // parent's, so the ENOSPC warnings scrolled past in CI without the script ever seeing
+  // them. spawnSync hands both streams back.
+  const run = spawnSync(
+    "npm",
+    ["install", "-g", "--prefix", prefix, "--no-audit", "--no-fund", spec],
+    { encoding: "utf8", maxBuffer: 512 * 1024 * 1024 }
+  );
+  const output = `${run.stdout ?? ""}${run.stderr ?? ""}`;
+  // Keep the install log visible, but a truncated package emits thousands of identical
+  // warnings — collapse them so the real message is not buried.
+  const stderrLines = String(run.stderr ?? "").split("\n");
+  const shown = stderrLines.length > 60 ? stderrLines.slice(0, 40) : stderrLines;
+  if (String(run.stderr ?? "").trim()) {
+    process.stderr.write(shown.join("\n") + "\n");
+    if (stderrLines.length > 60) {
+      process.stderr.write(`[install-upgrade] … ${stderrLines.length - 40} more npm line(s)\n`);
+    }
+  }
+  assertNoDiskExhaustion(output, label);
+  if (run.error) throw run.error;
+  if (run.status !== 0) {
+    throw new Error(`${label}: npm install exited with code ${run.status}`);
+  }
 }

+export function assertNoDiskExhaustion(output, label) {
+  if (!/ENOSPC|no space left on device/i.test(output)) return;
+  const count = (output.match(/ENOSPC/g) ?? []).length;
+  throw new Error(
+    `${label}: the install ran out of disk space (${count} ENOSPC error(s) from npm). ` +
+      `The package tree is truncated, so anything measured from it — boot, schema, ` +
+      `migrations — is meaningless. Free space in ${workDirForMessages} (each install tree is ` +
+      `~3 GB) and re-run. This is an environment failure, NOT a schema divergence.`
+  );
+}
+
+/** Best-effort free bytes on the filesystem backing `dir`, or null when unavailable. */
+function freeBytes(dir) {
+  try {
+    return fs.statfsSync(dir).bavail * fs.statfsSync(dir).bsize;
+  } catch {
+    return null;
+  }
+}
+
+const GB = 1024 ** 3;
+
+// Set once the work directory exists, so the ENOSPC message names the filesystem that
+// actually ran out — pointing at /tmp when the gate works elsewhere sends the reader to
+// free space on the wrong volume (which is what happened during the v3.8.50 publish).
+let workDirForMessages = os.tmpdir();
+
 function resolvePreviousVersion(current, explicit) {
  if (explicit) return explicit;
  const out = execFileSync("npm", ["view", "omniroute", "dist-tags.latest"], { encoding: "utf8" });
@@ -198,7 +331,9 @@ function resolvePreviousVersion(current, explicit) {
  if (latest === current) {
    // The version under test is already published (re-run of a shipped release): step back
    // to the highest published version strictly below it.
-    const all = JSON.parse(execFileSync("npm", ["view", "omniroute", "versions", "--json"], { encoding: "utf8" }));
+    const all = JSON.parse(
+      execFileSync("npm", ["view", "omniroute", "versions", "--json"], { encoding: "utf8" })
+    );
    const stable = all.filter((v) => !/-(rc|alpha|beta|pre|next)/.test(v) && v !== current);
    return stable[stable.length - 1];
  }
@@ -218,11 +353,25 @@ async function main() {
  }
  const version = JSON.parse(fs.readFileSync(path.join(ROOT, "package.json"), "utf8")).version;
  const allowlist = loadAllowlist(ROOT);
-  const tmp = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-install-upgrade-"));
+  // NOT os.tmpdir(): on the self-hosted runner /tmp is a 12 GB tmpfs backed by RAM, while
+  // the root filesystem has ~66 GB free. This gate needs ~12 GB, so it exhausted the tmpfs
+  // and npm truncated the package — 58269 ENOSPC errors on the v3.8.50 publish, which the
+  // previous code could only report as a crash. Freeing disk did not help because the disk
+  // was never the constraint. Work on real disk beside the repo instead.
+  const workRoot =
+    process.env.OMNIROUTE_INSTALL_UPGRADE_WORKDIR || path.join(ROOT, ".install-upgrade");
+  fs.mkdirSync(workRoot, { recursive: true });
+  const tmp = fs.mkdtempSync(path.join(workRoot, "omniroute-install-upgrade-"));
+  workDirForMessages = tmp;
  const failures = [];
  const warnings = [];

  try {
+    // Timed, because this turned out to be the expensive part: on the 2026-08-27
+    // v3.8.50 publish `npm pack` alone took 24m37s, leaving 5 of the step's 30-minute
+    // budget for two installs and two boots. Without a duration here the log showed
+    // only "packing…" then a timeout, which reads like a hang and is not.
+    const packStarted = Date.now();
    log(`packing v${version}…`);
    const packOut = execFileSync("npm", ["pack", "--json", "--pack-destination", tmp], {
      cwd: ROOT,
@@ -230,13 +379,31 @@ async function main() {
      maxBuffer: 128 * 1024 * 1024,
    });
    const tarball = path.join(tmp, pickTarball(packOut));
+    const packMb = (fs.statSync(tarball).size / 1024 / 1024).toFixed(1);
+    log(`packed in ${Math.round((Date.now() - packStarted) / 1000)}s (${packMb} MB)`);
+
+    // Each install tree is ~3 GB and this run builds two of them, side by side, plus the
+    // ~275 MB tarball. On the v3.8.50 publish run that overflowed the runner disk mid-way
+    // through the Phase B upgrade install; npm warned per truncated tar entry and still
+    // exited 0, and every later measurement was taken from a broken tree.
+    const availableBytes = freeBytes(tmp);
+    if (availableBytes !== null) {
+      log(`free space in ${tmp}: ${(availableBytes / GB).toFixed(1)} GB`);
+      if (availableBytes < 12 * GB) {
+        warn(
+          `only ${(availableBytes / GB).toFixed(1)} GB free — this gate needs roughly 12 GB ` +
+            `(two ~3 GB install trees, the second installed over twice, plus the tarball). ` +
+            `An install truncated by ENOSPC looks like a schema divergence.`
+        );
+      }
+    }

    // ---- Phase A: clean install -------------------------------------------------
    log("PHASE A — clean install of the packed tarball");
    const aPrefix = path.join(tmp, "a-prefix");
    const aData = path.join(tmp, "a-data");
    fs.mkdirSync(aData, { recursive: true });
-    npmInstallInto(aPrefix, tarball);
+    npmInstallInto(aPrefix, tarball, "clean install");
    const a = await bootAndProbe({
      prefix: aPrefix,
      dataDir: aData,
@@ -245,14 +412,34 @@ async function main() {
      label: "clean",
    });
    failures.push(...a.failures);
+    const cleanBooted = a.ok;
    if (a.ok) log(`clean install healthy on v${a.version}`);
+    else if (a.tail?.length) {
+      console.error("[install-upgrade] last output from the clean-install server:");
+      console.error(a.tail.join("").split("\n").slice(-40).join("\n"));
+    }
    const aDb = findDb(aData);
    const freshTables = aDb ? readTables(aDb) : null;
    if (!freshTables) failures.push("clean: no SQLite database was created");
    else log(`clean install schema: ${freshTables.size} tables`);

+    // Phase A is fully measured (boot verdict + schema snapshot); its ~3 GB install tree is
+    // dead weight from here on and Phase B needs the room. The DATA_DIR stays — only the
+    // node_modules tree goes.
+    if (!skipUpgrade) {
+      fs.rmSync(aPrefix, { recursive: true, force: true });
+      const reclaimed = freeBytes(tmp);
+      log(
+        "released the clean-install tree before the upgrade phase" +
+          (reclaimed !== null ? ` (${(reclaimed / GB).toFixed(1)} GB free)` : "")
+      );
+    }
+
    // ---- Phase B: upgrade over the previous published version -------------------
    let upgradedTables = null;
+    // `--skip-upgrade` never reaches the convergence block (upgradedTables stays null), so
+    // defaulting this to true keeps that path unchanged.
+    let upgradeBooted = true;
    if (skipUpgrade) {
      warn("PHASE B skipped (--skip-upgrade)");
    } else {
@@ -262,7 +449,7 @@ async function main() {
      const bData = path.join(tmp, "b-data");
      fs.mkdirSync(bData, { recursive: true });

-      npmInstallInto(bPrefix, `omniroute@${previous}`);
+      npmInstallInto(bPrefix, `omniroute@${previous}`, `previous(${previous}) install`);
      const before = await bootAndProbe({
        prefix: bPrefix,
        dataDir: bData,
@@ -273,14 +460,16 @@ async function main() {
      if (!before.ok) {
        // A broken PREVIOUS version is not this release's fault — degrade to a warning so a
        // historically bad publish cannot block the current one.
-        warnings.push(`previous version ${previous} did not boot cleanly — upgrade path unverified`);
+        warnings.push(
+          `previous version ${previous} did not boot cleanly — upgrade path unverified`
+        );
        for (const f of before.failures) warn(f);
      } else {
        const beforeDb = findDb(bData);
        const beforeTables = beforeDb ? readTables(beforeDb) : new Set();
        log(`previous(${previous}) schema: ${beforeTables.size} tables — upgrading in place`);

-        npmInstallInto(bPrefix, tarball);
+        npmInstallInto(bPrefix, tarball, "upgrade install");
        const after = await bootAndProbe({
          prefix: bPrefix,
          dataDir: bData,
@@ -290,6 +479,11 @@ async function main() {
        });
        failures.push(...after.failures);
        if (after.ok) log(`upgrade healthy on v${after.version}`);
+        upgradeBooted = after.ok;
+        if (!after.ok && after.tail?.length) {
+          console.error("[install-upgrade] last output from the upgraded server:");
+          console.error(after.tail.join("").split("\n").slice(-40).join("\n"));
+        }

        const afterDb = findDb(bData);
        upgradedTables = afterDb ? readTables(afterDb) : null;
@@ -306,7 +500,19 @@ async function main() {
    }

    // ---- Schema convergence -----------------------------------------------------
-    if (freshTables && upgradedTables) {
+    // Only meaningful when BOTH servers actually served. A boot that died before serving
+    // never ran a migration, so its database still holds the PREVIOUS release's schema and
+    // every post-baseline table shows up as "a clean install creates but an upgrade does
+    // not" — which is what the v3.8.50 publish run reported after ENOSPC truncated the
+    // upgrade install. Comparing there does not add information, it manufactures a
+    // 15-table false alarm on top of the real failure. The run still fails: the boot
+    // failure is already in `failures`.
+    if (freshTables && upgradedTables && !(cleanBooted && upgradeBooted)) {
+      warn(
+        "schema convergence NOT evaluated — a phase failed to boot, so its database was " +
+          "never migrated and any table difference would describe the broken boot, not the schema"
+      );
+    } else if (freshTables && upgradedTables) {
      const verdict = evaluateConvergence({
        freshTables,
        upgradedTables,
@@ -334,7 +540,10 @@ async function main() {

 // Only run the (expensive) gate when invoked directly — importing this module for the pure
 // helper above must not pack, install or boot anything.
-if (process.argv[1] && path.resolve(process.argv[1]) === path.resolve(new URL(import.meta.url).pathname)) {
+if (
+  process.argv[1] &&
+  path.resolve(process.argv[1]) === path.resolve(new URL(import.meta.url).pathname)
+) {
  main().catch((err) => {
    console.error(`[install-upgrade] crashed: ${err?.message ?? err}`);
    process.exit(1);
--- a/scripts/check/check-workflows.mjs
+++ b/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 — ${total} workflow finding(s) total (--strict mode).`);
--- a/scripts/check/lib/configExpiry.mjs
+++ b/scripts/check/lib/configExpiry.mjs
@@ -0,0 +1,90 @@
+/**
+ * scripts/check/lib/configExpiry.mjs
+ *
+ * Finds dated validity fields in JSON config packs so a test can fail BEFORE
+ * they lapse. Origin: config/alibaba-free-tier-allowlist.json carried
+ * `"validUntil": "2026-08-27"`; on 2026-08-28 the loader started (correctly)
+ * rejecting the pack and a test that asserted "the shipped pack loads" turned
+ * every PR and main red with no commit involved (#11866). A time bomb, not a
+ * regression — and the only kind of defect a diff review can never catch.
+ *
+ * Pure helpers; the repo-wide assertion lives in
+ * tests/unit/config-expiry-time-bomb.test.ts.
+ */
+import fs from "node:fs";
+import path from "node:path";
+
+export const EXPIRY_KEY =
+  /^(validUntil|valid_until|validTo|valid_to|expiresAt|expires_at|expiry|expires)$/;
+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("/"));
+  });
+}
--- a/scripts/check/lib/provenanceRunner.mjs
+++ b/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.`;
+}
--- a/scripts/ops/runner-janitor.sh
+++ b/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"
--- a/src/lib/db/migrations/163_model_capabilities.sql
+++ b/src/lib/db/migrations/163_model_capabilities.sql
@@ -0,0 +1,43 @@
+-- 163_model_capabilities.sql
+--
+-- Promote `model_capabilities` from a lazily-created runtime table to a real migration.
+--
+-- WHY: the table was only ever created by `ensureCapabilitiesTable()` in
+-- src/lib/modelsDevSync.ts, on demand, the first time a models.dev capability sync ran.
+-- Whether a database has it therefore depends on TIMING, not on the schema version — so a
+-- clean install and an upgraded install diverge for no structural reason. The v3.8.50
+-- publish run hit exactly that: `check:install-upgrade` reported `model_capabilities` as a
+-- table "present only after upgrade", because the older database had already run a sync
+-- and the freshly-installed one had not.
+--
+-- Creating it here makes both install paths converge deterministically.
+-- `ensureCapabilitiesTable()` stays in place as an idempotent safety net (it is a
+-- CREATE TABLE IF NOT EXISTS and now always a no-op); tests/unit/db-install-upgrade-schema-parity.test.ts
+-- pins the two definitions against drift.
+--
+-- IF NOT EXISTS is required, not decorative: every database that ever ran a models.dev
+-- sync already has this table, and this migration must be a no-op there.
+
+CREATE TABLE IF NOT EXISTS model_capabilities (
+  provider TEXT NOT NULL,
+  model_id TEXT NOT NULL,
+  tool_call BOOLEAN,
+  reasoning BOOLEAN,
+  attachment BOOLEAN,
+  structured_output BOOLEAN,
+  temperature BOOLEAN,
+  modalities_input TEXT,
+  modalities_output TEXT,
+  knowledge_cutoff TEXT,
+  release_date TEXT,
+  last_updated TEXT,
+  status TEXT,
+  family TEXT,
+  open_weights BOOLEAN,
+  limit_context INTEGER,
+  limit_input INTEGER,
+  limit_output INTEGER,
+  interleaved_field TEXT,
+  last_synced TEXT,
+  PRIMARY KEY (provider, model_id)
+);
--- a/tests/unit/alibaba-free-tier-allowlist.test.ts
+++ b/tests/unit/alibaba-free-tier-allowlist.test.ts
@@ -7,6 +7,9 @@
 */
 import { test } from "node:test";
 import assert from "node:assert/strict";
+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,
@@ -27,18 +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 packPath = `${process.cwd()}/config/alibaba-free-tier-allowlist.json`;
  process.env.ALIBABA_FREE_TIER_ALLOWLIST_PATH = packPath;
  resetAlibabaFreeTierAllowlistCache();
+  try {
+    assertions();
+  } finally {
+    if (previousPath) process.env.ALIBABA_FREE_TIER_ALLOWLIST_PATH = previousPath;
+    else delete process.env.ALIBABA_FREE_TIER_ALLOWLIST_PATH;
+    resetAlibabaFreeTierAllowlistCache();
+    rmSync(dir, { recursive: true, force: true });
+  }
+}

-  const pack = loadAlibabaFreeTierAllowlistPack();
-  assert.ok(pack);
-  assert.ok(isAlibabaFreeTierAllowlistPackValid(pack!));
-  assert.ok(pack!.capable.includes("qwen3.6-plus"));
+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);
+    }
+  );
+});

-  if (previousPath) process.env.ALIBABA_FREE_TIER_ALLOWLIST_PATH = previousPath;
-  else delete process.env.ALIBABA_FREE_TIER_ALLOWLIST_PATH;
-  resetAlibabaFreeTierAllowlistCache();
+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
+  );
 });
--- a/tests/unit/check-install-upgrade-convergence.test.ts
+++ b/tests/unit/check-install-upgrade-convergence.test.ts
@@ -2,7 +2,10 @@ import assert from "node:assert/strict";
 import test from "node:test";

 // @ts-expect-error — plain .mjs gate script, no type declarations by design
-import { evaluateConvergence } from "../../scripts/check/check-install-upgrade.mjs";
+import {
+  assertNoDiskExhaustion,
+  evaluateConvergence,
+} from "../../scripts/check/check-install-upgrade.mjs";

 /**
 * The whole point of this gate is that the two directions of schema divergence are NOT
@@ -66,7 +69,11 @@ test("UNKNOWN residue fails — a new divergence must not hide behind the allowl
  assert.equal(v.ok, false);
  assert.deepEqual(v.unknownResidue, ["surprise_table"]);
  assert.match(v.failures[0], /surprise_table/);
-  assert.doesNotMatch(v.failures[0], /cache_metrics/, "the known one must not be re-reported as new");
+  assert.doesNotMatch(
+    v.failures[0],
+    /cache_metrics/,
+    "the known one must not be re-reported as new"
+  );
 });

 test("both directions at once report both failures", () => {
@@ -93,3 +100,43 @@ test("empty/missing inputs do not crash", () => {
  assert.equal(v.ok, true);
  assert.deepEqual(v.onlyFresh, []);
 });
+
+// ─── ENOSPC guard ──────────────────────────────────────────────────────────────
+//
+// The v3.8.50 publish run (CI 33104507735) failed with "15 tables a CLEAN install creates
+// but an UPGRADE does not". None of them was missing: the Phase B upgrade install had hit
+// `npm warn tar TAR_ENTRY_ERROR ENOSPC: no space left on device` 5611 times, npm still
+// exited 0, the truncated `omniroute serve` "exited with code 0 before serving", and the
+// database therefore still held the 3.8.49 schema. npm reporting disk exhaustion as a
+// warning is what let a full disk masquerade as a schema defect.
+
+test("npm ENOSPC warnings are raised as an install failure, not ignored", () => {
+  const enospc = "npm warn tar TAR_ENTRY_ERROR ENOSPC: no space left on device, write\n".repeat(3);
+  assert.throws(
+    () => assertNoDiskExhaustion(enospc, "upgrade install"),
+    (err: Error) => {
+      assert.match(err.message, /upgrade install/);
+      assert.match(err.message, /ran out of disk space/);
+      assert.match(err.message, /3 ENOSPC error/);
+      // The operator must not go looking for a migration that is not missing.
+      assert.match(err.message, /NOT a schema divergence/);
+      return true;
+    }
+  );
+});
+
+test("a clean install log does not trip the disk guard", () => {
+  assert.doesNotThrow(() =>
+    assertNoDiskExhaustion(
+      "npm warn deprecated boolean@3.2.0: Package no longer supported.\nadded 900 packages\n",
+      "clean install"
+    )
+  );
+});
+
+test("the guard also catches the bare kernel message without the ENOSPC code", () => {
+  assert.throws(
+    () => assertNoDiskExhaustion("Error: no space left on device", "clean install"),
+    /ran out of disk space/
+  );
+});
--- a/tests/unit/check-workflows-provenance-runner.test.ts
+++ b/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)}`
+  );
+});
--- a/tests/unit/config-expiry-time-bomb.test.ts
+++ b/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"));
+});
--- a/tests/unit/db-install-upgrade-schema-parity.test.ts
+++ b/tests/unit/db-install-upgrade-schema-parity.test.ts
@@ -0,0 +1,118 @@
+// ENVIRONMENT NOTE (sandbox better-sqlite3 / glibc limitation, not a code defect):
+// This test opens a real SQLite database through `src/lib/db/core.ts`. better-sqlite3 is a
+// native addon; production and CI load it normally, but some sandboxes ship a system glibc
+// older than the prebuilt binary requires ("GLIBC_2.29 not found"), in which case the
+// runtime cascades to node:sqlite/sql.js. See tests/unit/_helpers/betterSqlite3Availability.ts.
+//
+// WHY THIS FILE EXISTS
+// --------------------
+// `npm run check:install-upgrade` (scripts/check/check-install-upgrade.mjs) proves that a
+// CLEAN install and an UPGRADE converge on the same schema, but it costs a full `npm pack`
+// plus three global installs and three boots (~17 min in CI) and it can only ever run at
+// publish time, against an already-published previous version. It is not a development
+// feedback loop, and the v3.8.50 publish run is what proved it: the gate reported 15
+// "missing" tables, and the deterministic half of that verdict was never checkable locally.
+//
+// This file pins the deterministic half in milliseconds:
+//
+//   1. Every migration file on disk is actually reachable by the runner on a fresh install.
+//      A file the runner never applies is a table no user ever gets.
+//   2. `model_capabilities` is created by the MIGRATION SET, not lazily at runtime.
+//      A table created on demand by `CREATE TABLE IF NOT EXISTS` inside a feature code
+//      path exists or not depending on whether that feature happened to run before the
+//      snapshot — so it diverges between the two install paths by TIMING, not by schema.
+//      That is precisely how `model_capabilities` surfaced as a divergence on the v3.8.50
+//      publish run (present in the upgraded database, absent from the clean one).
+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 { fileURLToPath } from "node:url";
+
+const REPO_ROOT = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..", "..");
+const MIGRATIONS_DIR = path.join(REPO_ROOT, "src", "lib", "db", "migrations");
+
+const TEST_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-install-upgrade-parity-"));
+process.env.DATA_DIR = TEST_DATA_DIR;
+process.env.DISABLE_SQLITE_AUTO_BACKUP = "true";
+
+const core = await import("../../src/lib/db/core.ts");
+
+// A clean install: core.ts applies the inline SCHEMA_SQL, the `ensure*Columns()` helpers,
+// then runMigrations(). This is the exact code path Phase A of the gate exercises.
+const db = core.getDbInstance();
+
+test.after(() => {
+  try {
+    core.resetDbInstance();
+  } catch {
+    /* best effort */
+  }
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+});
+
+function migrationFiles(): Array<{ version: string; name: string }> {
+  return fs
+    .readdirSync(MIGRATIONS_DIR)
+    .sort()
+    .map((file) => /^(\d{3,})_(.+)\.sql$/.exec(file))
+    .filter((m): m is RegExpExecArray => m !== null)
+    .map((m) => ({ version: m[1], name: m[2] }));
+}
+
+function hasTable(name: string): boolean {
+  return Boolean(
+    db.prepare("SELECT name FROM sqlite_master WHERE type = 'table' AND name = ?").get(name)
+  );
+}
+
+test("a clean install applies every migration file on disk", () => {
+  const ledger = new Set(
+    (
+      db.prepare("SELECT version FROM _omniroute_migrations").all() as Array<{ version: string }>
+    ).map((row) => row.version)
+  );
+  const unapplied = migrationFiles()
+    .filter((m) => !ledger.has(m.version))
+    .map((m) => `${m.version}_${m.name}`);
+  assert.deepEqual(
+    unapplied,
+    [],
+    "migration files the runner never applied — an upgrade would not create their tables either"
+  );
+});
+
+test("model_capabilities comes from the migration set, not from a lazy runtime CREATE", () => {
+  assert.ok(
+    hasTable("model_capabilities"),
+    "model_capabilities must be created by a migration so a clean install and an upgrade " +
+      "converge deterministically instead of depending on whether the models.dev sync ran"
+  );
+});
+
+test("the model_capabilities migration does not drift from ensureCapabilitiesTable()", () => {
+  const source = fs.readFileSync(path.join(REPO_ROOT, "src", "lib", "modelsDevSync.ts"), "utf8");
+  const ddl = /CREATE TABLE IF NOT EXISTS model_capabilities\s*\(([\s\S]*?)\n\s*\)/.exec(source);
+  assert.ok(ddl, "ensureCapabilitiesTable() DDL not found in src/lib/modelsDevSync.ts");
+
+  const runtimeColumns = ddl[1]
+    .split("\n")
+    .map((line) => line.trim())
+    .filter((line) => line.length > 0 && !/^PRIMARY KEY/i.test(line))
+    .map((line) => line.replace(/,$/, "").split(/\s+/)[0])
+    .sort();
+
+  const migrationColumns = (
+    db.prepare("PRAGMA table_info(model_capabilities)").all() as Array<{ name: string }>
+  )
+    .map((column) => column.name)
+    .sort();
+
+  assert.deepEqual(
+    migrationColumns,
+    runtimeColumns,
+    "the migration and the runtime helper must create the same columns — a drift here means " +
+      "an upgraded database keeps the old shape while a clean install gets the new one"
+  );
+});
--- a/tests/unit/npm-payload-nft-manifests-excluded.test.ts
+++ b/tests/unit/npm-payload-nft-manifests-excluded.test.ts
@@ -0,0 +1,72 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { readFileSync, readdirSync } from "node:fs";
+import type { Dirent } from "node:fs";
+import { join } from "node:path";
+
+/**
+ * v3.8.50 was refused by the registry with `413 Payload Too Large` on
+ * `POST /-/stage/package/omniroute`: the tarball had reached 288.7 MB packed
+ * (1.1 GB unpacked), against 174.5 MB for the 3.8.49 that published fine.
+ *
+ * 668.7 MB of that — 61% of the whole package — was 842 `*.nft.json` files.
+ * Those are Next.js Node File Trace manifests: build-time metadata used to
+ * COMPUTE the standalone bundle, never read while serving. They had doubled
+ * since 3.8.49 (325.0 MB across 748 files), which is what tipped the payload
+ * over the limit.
+ *
+ * The guard is the `files[]` negation, so a future entry that re-widens the
+ * glob (or a rewrite of the array) cannot silently put them back.
+ */
+const pkg = JSON.parse(readFileSync(join(import.meta.dirname, "../../package.json"), "utf8")) as {
+  files?: string[];
+};
+
+test("package.json files[] excludes Next's .nft.json trace manifests", () => {
+  const files = pkg.files ?? [];
+  assert.ok(files.length > 0, "package.json must declare files[]");
+  assert.ok(
+    files.includes("!**/*.nft.json"),
+    "files[] must negate **/*.nft.json — they are build metadata and were 61% of the 3.8.50 payload"
+  );
+});
+
+test("the negation sits after the positive dist/ entry it has to override", () => {
+  // npm applies files[] in order: a negation listed BEFORE the directory that
+  // pulls the files in is a no-op. Positive anchor, so this test cannot pass
+  // just because both strings happen to be present somewhere.
+  const files = pkg.files ?? [];
+  const dist = files.indexOf("dist/");
+  const negation = files.indexOf("!**/*.nft.json");
+  assert.notEqual(dist, -1, "dist/ must still be published");
+  assert.ok(negation > dist, "the .nft.json negation must come after dist/");
+});
+
+test("no source module reads a .nft.json at runtime", () => {
+  // If this ever stops holding, the exclusion above becomes a runtime break
+  // rather than a size win — which is exactly the assumption worth pinning.
+  const roots = ["src", "open-sse", "bin"];
+  const hits: string[] = [];
+  for (const root of roots) {
+    const dir = join(import.meta.dirname, "../..", root);
+    const stack = [dir];
+    while (stack.length > 0) {
+      const current = stack.pop() as string;
+      let entries: Dirent[];
+      try {
+        entries = readdirSync(current, { withFileTypes: true });
+      } catch {
+        continue;
+      }
+      for (const entry of entries) {
+        const full = join(current, entry.name);
+        if (entry.isDirectory()) {
+          if (entry.name !== "node_modules") stack.push(full);
+        } else if (/\.(ts|tsx|mjs|js)$/.test(entry.name)) {
+          if (readFileSync(full, "utf8").includes(".nft.json")) hits.push(full);
+        }
+      }
+    }
+  }
+  assert.deepEqual(hits, [], `nothing may depend on .nft.json at runtime: ${hits.join(", ")}`);
+});
--- a/tests/unit/runner-janitor.test.ts
+++ b/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);
+  });
+});
				`@@ -0,0 +1 @@`
				- Electron release: `electron/package-lock.json` regained the optional `electron-builder-squirrel-windows` subtree (13 entries) that `npm ci` had been refusing as out of sync — the Linux desktop leg died on it — and `electron-release.yml` gained a `build_ref` dispatch input so a release whose tag was cut with the broken lock can have its assets rebuilt from the repaired line
				`@@ -0,0 +1 @@`
				- Electron release workflow: the `publish-npm` job now grants `actions: read` to the reusable `npm-publish.yml` it calls (its `publish` job requests it), which is what made GitHub refuse the whole v3.8.50 run at startup and ship the release with zero desktop assets; a `workflow_dispatch` now builds the requested tag instead of the dispatching branch and can skip the npm leg (`publish_npm=false`) when only re-attaching assets
				`@@ -0,0 +1 @@`
				- npm publish workflow: the CycloneDX SBOM is attached to the GitHub Release on `workflow_dispatch` publishes too (when a release for the tag exists), not only on the `release` event — v3.8.50 shipped through a staged dispatch and its release carried no SBOM until it was attached by hand from the run's `sbom-npm` artifact
				`@@ -0,0 +1 @@`
				- `Coverage` job on `ci.yml`: the informational Codecov upload gets its own 5-minute ceiling and `continue-on-error`, and the job budget grows from 20 to 30 minutes (the 8-shard c8 merge alone takes ~10) — a stalled upload no longer ends the job `cancelled` and drags a fully green `main` run's conclusion down with it