ci: lockstep HEAL — baked-tools.json auto-syncs when Dockerfile pin leads (#323)

.github/workflows/baked-tools-check.yml

@@ -28,6 +28,10 @@ jobs:
     name: Validate baked tools in image
     runs-on: ubuntu-latest
     timeout-minutes: 20
+    # #323: lockstep-heal needs contents:write to push the sync commit to the PR head
+    permissions:
+      contents: write
+      checks: read
 
     steps:
       - name: Checkout builder repository
@@ -37,6 +41,10 @@ jobs:
           fetch-depth: 1
 
       - name: Manifest <-> Dockerfile lockstep gate
+        id: gate
+        # #323: let the heal step run on mismatch; a hard-fail step would stop the job here.
+        # The 'Fail if still red' step below re-asserts the red when healing is impossible.
+        continue-on-error: true
         run: |
           python3 - << 'GATE'
           import json, re, sys
@@ -81,6 +89,100 @@ jobs:
           print('Lockstep OK: all Dockerfile pins match baked-tools.json')
           GATE
 
+      # #323 lockstep-heal: when the ONLY failures are 'Dockerfile pin X != baked-tools.json Y'
+      # (Dockerfile leads — dependabot FROM bumps, manual @getpaseo bumps) and the PR did
+      # NOT hand-edit baked-tools.json, sync baked-tools.json to the Dockerfile pins and
+      # push to the PR head. This run then fails intentionally ('HEALED — rerun verifies');
+      # the pushed commit re-triggers the checks and auto-merge proceeds on the new SHA.
+      # Never heals the reverse direction (baked-tools.json leading) — that needs a human.
+      - name: Checkout PR head (for heal)
+        id: healsrc
+        if: steps.gate.outcome == 'failure'
+        uses: actions/checkout@v7
+        with:
+          ref: ${{ github.event.pull_request.head.ref }}
+          path: healsrc
+          fetch-depth: 1
+
+      - name: Heal lockstep drift (Dockerfile pin leads)
+        id: heal
+        if: steps.gate.outcome == 'failure'
+        run: |
+          set -euo pipefail
+          if [ "${{ github.event_name }}" != "pull_request" ] \
+             || [ "${{ github.event.pull_request.head.repo.full_name }}" != "${{ github.repository }}" ]; then
+            echo "not healable (fork PR or non-PR context) — leaving red for a human"
+            exit 0
+          fi
+          HEAD_BRANCH="${{ github.event.pull_request.head.ref }}"
+          if gh api "repos/${{ github.repository }}/pulls/${{ github.event.pull_request.number }}/files" \
+              --paginate --jq '.[].filename' 2>/dev/null | grep -qx 'managed-tools/baked-tools.json'; then
+            echo "PR touches baked-tools.json — heal disabled (human edit in flight)"
+            exit 0
+          fi
+          cd healsrc
+          set +e
+          python3 - <<'HEAL'
+          import json, re, sys
+          df = open('Dockerfile.dockerfile').read()
+          path = 'managed-tools/baked-tools.json'
+          data = json.load(open(path))
+          manifest = {t['name']: t for t in data['tools']}
+          pins = {
+              'docker':      (r'FROM docker:([0-9][\w.]*)-dind', None),
+              'dockerd':     (r'FROM docker:([0-9][\w.]*)-dind', None),
+              'paseo':       (r'@getpaseo/cli@([0-9][\w.]*)', r'@getpaseo/server@([0-9][\w.]*)'),
+              'paseo-relay': (r'ARG PASEO_RELAY_VERSION=([0-9][\w.]*)', None),
+              'code-server': (r'ARG CODE_SERVER_VERSION=([0-9][\w.]*)', None),
+          }
+          mismatch = re.compile(r'^(\S+): (?:secondary )?Dockerfile pin ([0-9][\w.]*) != baked-tools\.json ([0-9][\w.]*)$')
+          healed = []
+          only_mismatches = True
+          # recompute failures exactly like the gate, but capture them for classification
+          for tool, (rx, rx2) in pins.items():
+              if tool not in manifest:
+                  only_mismatches = False; continue
+              want = manifest[tool]['currentVersion']
+              got = re.search(rx, df)
+              if not got:
+                  only_mismatches = False; continue
+              if got.group(1) != want:
+                  healed.append((tool, got.group(1), want))
+              if rx2:
+                  got2 = re.search(rx2, df)
+                  if got2 and got2.group(1) != want:
+                      pass  # same tool, already captured above
+          if not only_mismatches or not healed:
+              print('no healable drift (non-mismatch failures or clean) — leaving red')
+              sys.exit(1)
+          for tool, new, old in healed:
+              manifest[tool]['currentVersion'] = new
+              print(f'healed {tool}: {old} -> {new} (Dockerfile leads)')
+          json.dump(data, open(path, 'w'), indent=2)
+          open(path, 'a').write('\n')
+          sys.exit(42)
+          HEAL
+          rc=$?
+          set -e
+          if [ "$rc" -ne 42 ]; then
+            echo "heal: nothing pushed (rc=$rc)"
+            exit 0
+          fi
+          git config user.name "lockstep-heal[bot]"
+          git config user.email "action@github.com"
+          git add managed-tools/baked-tools.json
+          git commit -q -m "chore(baked-tools): lockstep heal — sync baked-tools.json to Dockerfile pins (automated, #323)"
+          git push origin "HEAD:refs/heads/$HEAD_BRANCH"
+          echo "pushed=true" >> "$GITHUB_OUTPUT"
+          echo "HEALED — sync commit pushed; this run fails intentionally so nothing merges on the pre-heal SHA."
+          exit 1
+
+      - name: Fail if still red and not healed
+        if: steps.gate.outcome == 'failure' && steps.heal.outputs.pushed != 'true'
+        run: |
+          echo "Lockstep gate failed and healing did not apply — see gate output above."
+          exit 1
+
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v4