Upload .github/workflows/librarian_healthcheck.yml with huggingface_hub

.github/workflows/librarian_healthcheck.yml

@@ -0,0 +1,97 @@
+name: Librarian Healthcheck
+
+# Polls the deployed HF Space's /v1/system/status endpoint on a schedule.
+# If the Space is unhealthy (non-2xx, missing fields, or unreachable) the
+# job triggers `hf-sync.yml` so the Space is rebuilt from the latest main.
+#
+# Self-healing per Phase 3.3 of UNIVERSAL_ALIGNMENT_PROTOCOL_QGTNL.
+
+on:
+  schedule:
+    # Every 6 hours, on the hour.
+    - cron: '0 */6 * * *'
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  actions: write   # required to dispatch hf-sync.yml on failure
+
+concurrency:
+  group: librarian-healthcheck
+  cancel-in-progress: false
+
+jobs:
+  ping-status:
+    runs-on: ubuntu-latest
+    env:
+      HF_SPACE: DJ-Goanna-Coding/Oppo-TIA
+      STATUS_URL: https://dj-goanna-coding-oppo-tia.hf.space/v1/system/status
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 1
+
+      - name: Probe /v1/system/status
+        id: probe
+        run: |
+          set +e
+          # Allow the Space up to 90s to wake up if it was sleeping.
+          response=$(curl -fsS --max-time 90 --retry 3 --retry-delay 10 \
+            -H "Accept: application/json" "$STATUS_URL")
+          rc=$?
+          set -e
+
+          if [ $rc -ne 0 ]; then
+            echo "::warning::Status endpoint unreachable (curl rc=$rc)"
+            echo "healthy=false" >> "$GITHUB_OUTPUT"
+            echo "reason=unreachable" >> "$GITHUB_OUTPUT"
+            exit 0
+          fi
+
+          echo "$response" | python3 -c "
+          import json, sys
+          payload = json.load(sys.stdin)
+          required = {'repo', 'version', 'commit', 'index_built', 'personas', 'uptime_seconds'}
+          missing = required - payload.keys()
+          if missing:
+              print(f'::warning::status payload missing fields: {sorted(missing)}')
+              sys.exit(2)
+          print(f\"OK  repo={payload['repo']}  commit={payload['commit']}  \"
+                f\"personas={len(payload['personas'])}  \"
+                f\"index_built={payload['index_built']}  \"
+                f\"uptime={payload['uptime_seconds']}s\")
+          "
+          rc=$?
+          if [ $rc -eq 0 ]; then
+            echo "healthy=true" >> "$GITHUB_OUTPUT"
+          else
+            echo "healthy=false" >> "$GITHUB_OUTPUT"
+            echo "reason=schema-mismatch" >> "$GITHUB_OUTPUT"
+          fi
+
+      - name: Trigger hf-sync.yml on failure
+        if: steps.probe.outputs.healthy != 'true'
+        env:
+          GH_TOKEN: ${{ github.token }}
+          REASON: ${{ steps.probe.outputs.reason }}
+        run: |
+          echo "Space unhealthy (${REASON}); dispatching hf-sync.yml to rebuild..."
+          gh workflow run hf-sync.yml \
+            --repo "${{ github.repository }}" \
+            --ref "${{ github.ref_name }}"
+          echo "Dispatched."
+
+      - name: Emit librarian action log
+        if: always()
+        env:
+          STATUS: ${{ steps.probe.outputs.healthy == 'true' && 'ok' || 'fail' }}
+          REASON: ${{ steps.probe.outputs.reason }}
+        run: |
+          # Best-effort: never fail the job because of logging.
+          python3 scripts/emit_action_log.py \
+            --action librarian-healthcheck \
+            --actor github-actions \
+            --status "${STATUS}" \
+            --detail "{\"reason\": \"${REASON:-none}\", \"space\": \"${HF_SPACE}\"}" \
+            --quiet || true