ci: quality gates phase 1 -- dupl repo-wide, gocognit clean-as-you-code (goal 0109)

.claude/rules/testing.md

@@ -294,6 +294,32 @@ folklore in agent-brief prose. The standing rules:
   assertions); octocov-style coverage actions (a 15-line floor
   script suffices; no new CI dependency).
 
+## Quality gates: duplication + cognitive complexity (goal 0109)
+
+The SonarQube-class gates, wired measurement-first (a full read-only
+run over the tree BEFORE thresholds were committed — hit lists in the
+goal file), never aspiration-first:
+
+- **Duplication (`dupl` @ 150, repo-wide)**: lands green because the
+  only existing clusters are deliberate and excluded BY NAME — test
+  twins (`_test.go`), the per-entity seed/CRUD shape in
+  `configuresvc/`, and `atlasservice_builtin.go`. New duplication
+  anywhere else fails the build. Anti-gaming: exclusions are named
+  files/families with recorded reasons, never a threshold raise.
+- **Cognitive complexity (`gocognit` @ 15, NEW/CHANGED code only)**:
+  25 legacy production functions sit over the threshold (max 65 —
+  burn-down list in the goal file), so the gate runs Sonar's own
+  clean-as-you-code posture via `--enable-only gocognit
+  --new-from-merge-base=origin/main` (lefthook `gocognit-new` + a CI
+  step; the lint job's checkout carries `fetch-depth: 0` for the
+  merge-base). Legacy refuses to rot further without failing today's
+  build; touching a legacy offender means paying its complexity down
+  or consciously splitting the change.
+- **Deferred to the npm-deps landing** (lockfile-conflict avoidance,
+  recorded in the goal file): eslint-plugin-sonarjs (TS cognitive
+  complexity + duplicate-branch rules) and the diff-cover
+  changed-lines coverage gate.
+
 ## Shared-pool vs dedicated e2e servers — declare it up front
 
 A spec whose assertions read GLOBAL app state that other tests can

.github/workflows/ci.yml

@@ -206,6 +206,11 @@ jobs:
       CGO_ENABLED: '0'
     steps:
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          # Full history: the gocognit new-code gate below diffs against
+          # the merge-base with origin/main, which a depth-1 clone
+          # cannot compute.
+          fetch-depth: 0
       - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
         with:
           name: frontend-dist
@@ -218,6 +223,12 @@ jobs:
         with:
           version: v2.12
           args: --build-tags=server
+      # Cognitive-complexity gate on NEW/CHANGED code only (goal 0109,
+      # the clean-as-you-code posture -- see lefthook.yml's gocognit-new
+      # job for the full reasoning). Reuses the binary the action above
+      # already installed onto PATH. Requires the full-history checkout
+      # this job's own fetch-depth: 0 provides (merge-base needs it).
+      - run: golangci-lint run . ./internal/... --build-tags=server --enable-only gocognit --new-from-merge-base=origin/main
 
   build-go:
     needs: [changes, frontend]
@@ -357,7 +368,29 @@ jobs:
           cache-dependency-path: frontend/package-lock.json
       - run: npm ci
         working-directory: frontend
-      - run: npx playwright install --with-deps chromium
+      # Browser binaries cached by Playwright version: the uncached
+      # `install --with-deps` path stalls in its apt phase on flaky
+      # ubuntu mirrors (three 15-minute shard hangs observed in one
+      # day), and the runner image already carries Chromium's system
+      # deps -- so a cache hit skips the whole install, and only a
+      # real Playwright version bump pays the download again (the
+      # upstream-documented CI caching pattern).
+      - name: Resolve Playwright version
+        id: pw-version
+        working-directory: frontend
+        run: echo "version=$(node -p "require('@playwright/test/package.json').version")" >> "$GITHUB_OUTPUT"
+      - uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
+        id: pw-cache
+        with:
+          path: ~/.cache/ms-playwright
+          key: playwright-chromium-${{ runner.os }}-${{ steps.pw-version.outputs.version }}
+      # No --with-deps: the ubuntu runner image already ships
+      # Chromium's system libraries, and the apt phase behind
+      # --with-deps is the exact stall (archive.ubuntu.com hanging
+      # until the job's 15-minute cap) this caching exists to kill --
+      # observed persisting even on cache-miss runs.
+      - if: steps.pw-cache.outputs.cache-hit != 'true'
+        run: npx playwright install chromium
         working-directory: frontend
       # 3-shard matrix + playwright.config.ts's workers: CI ? 1 : 4 (goal
       # 0024): the prior single-job, workers:2-in-CI shape was the actual

.golangci.yml

@@ -26,15 +26,48 @@ linters:
     - prealloc
     - contextcheck
     - sqlclosecheck
+    # Third pass (goal 0109, the SonarQube-class duplication gate):
+    # threshold and exclusions tuned from a full measurement run over
+    # the tree (28 pairs found, every one either the deliberate
+    # per-entity seed-reconcile shape or a test twin -- hit lists in
+    # the goal file), so the gate lands green and catches NEW
+    # duplication anywhere outside the named, deliberate families.
+    - dupl
   exclusions:
     paths:
       - frontend
       - build
       - third_party$
       - builtin$
       - examples$
+    rules:
+      # Test twins (a Go test asserting the same shape twice) are the
+      # documented-noise class Sonar's own defaults also exempt.
+      - path: _test\.go
+        linters: [dupl]
+      # The per-entity-kind seed/CRUD shape: each Configure entity
+      # repeats the same reconcile/upgrade body deliberately (goal
+      # 0109's measurement found every cluster here); collapsing them
+      # behind generics was considered and rejected -- the repetition
+      # is per-entity clarity, not an accident.
+      - path: internal/services/configuresvc/
+        linters: [dupl]
+      - path: internal/services/atlassvc/atlasservice_builtin\.go
+        linters: [dupl]
 
   settings:
+    dupl:
+      threshold: 150
+    # gocognit is deliberately NOT in `enable:` above -- 25 legacy
+    # production functions sit over its threshold (max 65; burn-down
+    # list in docs/goals/0109), so a repo-wide gate would either fail
+    # every build or need a threshold too high to mean anything.
+    # Instead it gates NEW/CHANGED code only (Sonar's own
+    # clean-as-you-code posture), via a dedicated
+    # `--enable-only gocognit --new-from-merge-base` invocation in
+    # lefthook and CI, which reads this threshold.
+    gocognit:
+      min-complexity: 15
     revive:
       rules:
         # This repo has never doc-commented every exported symbol --

lefthook.yml

@@ -56,6 +56,15 @@ pre-commit:
       # so a local pass here was checking something CI doesn't and a
       # local fail could block a commit CI's real gate would accept.
       run: golangci-lint run . ./internal/... --build-tags=server
+    - name: gocognit-new
+      glob: "*.go"
+      # Cognitive complexity gates NEW/CHANGED code only (goal 0109,
+      # Sonar's clean-as-you-code posture): 25 legacy functions sit
+      # over the threshold (burn-down in the goal file), so the gate
+      # scopes to the merge-base diff instead of failing every build
+      # or inflating the threshold into meaninglessness. Threshold
+      # lives in .golangci.yml's gocognit settings.
+      run: golangci-lint run . ./internal/... --build-tags=server --enable-only gocognit --new-from-merge-base=origin/main
     - name: go-test
       glob: "*.go"
       # Root package (`.`) included alongside ./internal/... -- the