Add weekly agent metrics aggregation

.github/workflows/agents-autofix-loop.yml

@@ -30,6 +30,11 @@ jobs:
       stop_reason: ${{ steps.evaluate.outputs.stop_reason }}
       attempts: ${{ steps.evaluate.outputs.attempts }}
       max_attempts: ${{ steps.evaluate.outputs.max_attempts }}
+      trigger_reason: ${{ steps.evaluate.outputs.trigger_reason }}
+      trigger_job: ${{ steps.evaluate.outputs.trigger_job }}
+      trigger_step: ${{ steps.evaluate.outputs.trigger_step }}
+      gate_conclusion: ${{ steps.evaluate.outputs.gate_conclusion }}
+      gate_run_id: ${{ steps.evaluate.outputs.gate_run_id }}
       security_blocked: ${{ steps.security_gate.outputs.blocked }}
       security_reason: ${{ steps.security_gate.outputs.reason }}
     steps:
@@ -110,6 +115,11 @@ jobs:
               stop_reason: '',
               attempts: '0',
               max_attempts: '3',
+              trigger_reason: 'unknown',
+              trigger_job: '',
+              trigger_step: '',
+              gate_conclusion: String(run?.conclusion || run?.status || ''),
+              gate_run_id: String(run?.id || ''),
             };
 
             const stop = (reason, stopReason = '') => {
@@ -202,6 +212,8 @@ jobs:
             outputs.max_attempts = String(maxAttempts);
 
             const failingJobs = [];
+            let triggerJob = null;
+            let triggerStep = null;
             for (const job of jobs) {
               const conclusion = (job.conclusion || job.status || '').toLowerCase();
               if (!conclusion || ['success', 'skipped'].includes(conclusion)) {
@@ -223,8 +235,36 @@ jobs:
                 detailLines.push(`  - steps: ${failingSteps.join('; ')}`);
               }
               failingJobs.push(detailLines.join('\n'));
+
+              if (!triggerJob) {
+                triggerJob = job;
+                const failingStep = Array.isArray(job.steps)
+                  ? job.steps.find((step) => {
+                      const stepConclusion = (step.conclusion || step.status || '').toLowerCase();
+                      return stepConclusion && !['success', 'skipped'].includes(stepConclusion);
+                    })
+                  : null;
+                triggerStep = failingStep || null;
+              }
             }
 
+            const inferTriggerReason = (job, step) => {
+              const text = [job?.name, step?.name]
+                .filter(Boolean)
+                .map((value) => String(value).toLowerCase())
+                .join(' ');
+
+              if (!text) return 'unknown';
+              if (text.includes('mypy')) return 'mypy';
+              if (text.includes('lint') || text.includes('flake8') || text.includes('ruff')) return 'lint';
+              if (text.includes('pytest') || text.includes('test')) return 'pytest';
+              return 'unknown';
+            };
+
+            outputs.trigger_reason = inferTriggerReason(triggerJob, triggerStep);
+            outputs.trigger_job = triggerJob?.name || triggerJob?.id || '';
+            outputs.trigger_step = triggerStep?.name || '';
+
             const appendixLines = [
               `Gate run: ${run.html_url || run.id}`,
               `Conclusion: ${run.conclusion || run.status || 'unknown'}`,
@@ -319,3 +359,140 @@ jobs:
               issue_number: prNumber,
               body,
             });
+
+  metrics:
+    name: Record autofix metrics
+    needs:
+      - prepare
+      - autofix
+    if: always()
+    runs-on: ubuntu-latest
+    environment: agent-standard
+    steps:
+      - name: Collect metrics
+        id: collect
+        uses: actions/github-script@v7
+        with:
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          script: |
+            const prNumber = Number('${{ needs.prepare.outputs.pr_number || 0 }}') || 0;
+            const attemptNumber = Number('${{ needs.prepare.outputs.attempts || 0 }}') || 0;
+            const attemptLimit = Number('${{ needs.prepare.outputs.max_attempts || 0 }}') || 0;
+            const headShaBefore = '${{ needs.prepare.outputs.head_sha }}';
+            const gateConclusionBefore = '${{ needs.prepare.outputs.gate_conclusion }}' || (context.payload.workflow_run?.conclusion || '');
+            const gateRunId = '${{ needs.prepare.outputs.gate_run_id }}' || String(context.payload.workflow_run?.id || '');
+            const triggerReason = '${{ needs.prepare.outputs.trigger_reason || 'unknown' }}';
+            const triggerJob = '${{ needs.prepare.outputs.trigger_job }}';
+            const triggerStep = '${{ needs.prepare.outputs.trigger_step }}';
+            const stopReason = '${{ needs.prepare.outputs.stop_reason }}';
+            const autofixResult = '${{ needs.autofix.result }}';
+
+            const { owner, repo } = context.repo;
+            let fixApplied = false;
+            let headShaAfter = headShaBefore;
+            let gateResultAfter = gateConclusionBefore || 'unknown';
+
+            if (prNumber) {
+              try {
+                const { data: pr } = await github.rest.pulls.get({
+                  owner,
+                  repo,
+                  pull_number: prNumber,
+                });
+                headShaAfter = pr.head?.sha || headShaAfter;
+                fixApplied = Boolean(headShaBefore && headShaAfter && headShaBefore !== headShaAfter);
+
+                const gateWorkflow = 'pr-00-gate.yml';
+                const runs = await github.paginate(github.rest.actions.listWorkflowRuns, {
+                  owner,
+                  repo,
+                  workflow_id: gateWorkflow,
+                  head_sha: headShaAfter,
+                  per_page: 20,
+                });
+                const latestGateRun = runs[0];
+                if (latestGateRun) {
+                  gateResultAfter = latestGateRun.conclusion || latestGateRun.status || 'unknown';
+                } else {
+                  gateResultAfter = 'not-found';
+                }
+              } catch (error) {
+                core.warning(`Failed to resolve PR or gate status: ${error.message}`);
+              }
+            }
+
+            const metrics = {
+              workflow_run_id: gateRunId,
+              pr_number: prNumber,
+              attempt_number: attemptNumber,
+              attempt_limit: attemptLimit,
+              trigger_reason: triggerReason || 'unknown',
+              trigger_job: triggerJob,
+              trigger_step: triggerStep,
+              fix_applied: fixApplied,
+              gate_result_after: gateResultAfter || 'unknown',
+              gate_conclusion_before: gateConclusionBefore || 'unknown',
+              stop_reason: stopReason || '',
+              autofix_result: autofixResult || 'unknown',
+              head_sha_before: headShaBefore,
+              head_sha_after: headShaAfter,
+              recorded_at: new Date().toISOString(),
+            };
+
+            core.setOutput('metrics_json', JSON.stringify(metrics));
+
+      - name: Write summary and artifact
+        env:
+          METRICS_JSON: ${{ steps.collect.outputs.metrics_json }}
+        run: |
+          set -euo pipefail
+          if [ -z "${METRICS_JSON:-}" ]; then
+            echo "No metrics JSON captured; skipping summary."
+            exit 0
+          fi
+
+          python - <<'PY'
+          import json
+          import os
+
+          metrics = json.loads(os.environ["METRICS_JSON"])
+          order = [
+              "pr_number",
+              "attempt_number",
+              "attempt_limit",
+              "trigger_reason",
+              "trigger_job",
+              "trigger_step",
+              "fix_applied",
+              "gate_conclusion_before",
+              "gate_result_after",
+              "autofix_result",
+              "stop_reason",
+              "workflow_run_id",
+              "head_sha_before",
+              "head_sha_after",
+              "recorded_at",
+          ]
+
+          lines = ["## Autofix loop metrics", ""] + ["| Field | Value |", "| --- | --- |"]
+          for key in order:
+              value = metrics.get(key, "")
+              lines.append(f"| {key} | `{value}` |")
+
+          summary_path = os.environ.get("GITHUB_STEP_SUMMARY")
+          if summary_path:
+              with open(summary_path, "a", encoding="utf-8") as fp:
+                  fp.write("\n".join(lines) + "\n")
+
+          out_path = "autofix-metrics.ndjson"
+          with open(out_path, "a", encoding="utf-8") as fp:
+              fp.write(json.dumps(metrics) + "\n")
+          print(f"Wrote metrics to {out_path}")
+          PY
+
+      - name: Upload metrics artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: agents-autofix-metrics
+          path: autofix-metrics.ndjson
+          retention-days: 30

.github/workflows/agents-keepalive-loop.yml

@@ -36,12 +36,16 @@ jobs:
       autofix_enabled: ${{ steps.evaluate.outputs.autofix_enabled }}
       has_agent_label: ${{ steps.evaluate.outputs.has_agent_label }}
       trace: ${{ steps.evaluate.outputs.trace }}
+      start_ts: ${{ steps.timestamps.outputs.start_ts }}
       security_blocked: ${{ steps.security_gate.outputs.blocked }}
       security_reason: ${{ steps.security_gate.outputs.reason }}
     steps:
       - name: Checkout
         uses: actions/checkout@v4
 
+      - name: Capture timestamps
+        id: timestamps
+        run: echo "start_ts=$(date -u +%s)" >> "$GITHUB_OUTPUT"
       - name: Security gate - prompt injection guard
         id: security_gate
         uses: actions/github-script@v7
@@ -146,6 +150,81 @@ jobs:
       - name: Checkout
         uses: actions/checkout@v4
 
+      - name: Emit keepalive metrics
+        id: keepalive-metrics
+        env:
+          PR_NUMBER: ${{ needs.evaluate.outputs.pr_number }}
+          ACTION: ${{ needs.evaluate.outputs.action }}
+          REASON: ${{ needs.evaluate.outputs.reason }}
+          GATE_CONCLUSION: ${{ needs.evaluate.outputs.gate_conclusion }}
+          ITERATION: ${{ needs.evaluate.outputs.iteration }}
+          MAX_ITERATIONS: ${{ needs.evaluate.outputs.max_iterations }}
+          TASKS_TOTAL: ${{ needs.evaluate.outputs.tasks_total }}
+          TASKS_UNCHECKED: ${{ needs.evaluate.outputs.tasks_unchecked }}
+          START_TS: ${{ needs.evaluate.outputs.start_ts }}
+        run: |
+          set -euo pipefail
+
+          now=$(date -u +%s)
+          if [[ "${START_TS:-}" =~ ^[0-9]+$ ]]; then
+            duration=$(( now - START_TS ))
+            if [ "$duration" -lt 0 ]; then duration=0; fi
+          else
+            duration=0
+          fi
+
+          tasks_total=${TASKS_TOTAL:-0}
+          tasks_unchecked=${TASKS_UNCHECKED:-0}
+          if ! [[ "$tasks_total" =~ ^-?[0-9]+$ ]]; then tasks_total=0; fi
+          if ! [[ "$tasks_unchecked" =~ ^-?[0-9]+$ ]]; then tasks_unchecked=0; fi
+          tasks_completed=$(( tasks_total - tasks_unchecked ))
+          if [ "$tasks_completed" -lt 0 ]; then tasks_completed=0; fi
+
+          metrics_json=$(jq -n \
+            --arg pr "${PR_NUMBER:-0}" \
+            --arg iteration "${ITERATION:-0}" \
+            --arg action "${ACTION:-}" \
+            --arg stop_reason "${REASON:-}" \
+            --arg gate_conclusion "${GATE_CONCLUSION:-}" \
+            --arg tasks_total "$tasks_total" \
+            --arg tasks_completed "$tasks_completed" \
+            --arg duration "$duration" \
+            '{
+              pr_number: ($pr | tonumber? // 0),
+              iteration_count: ($iteration | tonumber? // 0),
+              action: $action,
+              stop_reason: $stop_reason,
+              gate_conclusion: $gate_conclusion,
+              tasks_total: ($tasks_total | tonumber? // 0),
+              tasks_completed: ($tasks_completed | tonumber? // 0),
+              duration_seconds: ($duration | tonumber? // 0)
+            }')
+
+          {
+            echo '### Keepalive metrics'
+            echo ''
+            echo '| Field | Value |'
+            echo '| --- | --- |'
+            echo "| pr_number | $(echo "$metrics_json" | jq -r '.pr_number') |"
+            echo "| iteration_count | $(echo "$metrics_json" | jq -r '.iteration_count') |"
+            echo "| action | $(echo "$metrics_json" | jq -r '.action') |"
+            echo "| stop_reason | $(echo "$metrics_json" | jq -r '.stop_reason') |"
+            echo "| gate_conclusion | $(echo "$metrics_json" | jq -r '.gate_conclusion') |"
+            echo "| tasks_total | $(echo "$metrics_json" | jq -r '.tasks_total') |"
+            echo "| tasks_completed | $(echo "$metrics_json" | jq -r '.tasks_completed') |"
+            echo "| duration_seconds | $(echo "$metrics_json" | jq -r '.duration_seconds') |"
+          } >> "$GITHUB_STEP_SUMMARY"
+
+          echo "$metrics_json" >> keepalive-metrics.ndjson
+
+      - name: Upload keepalive metrics artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: keepalive-metrics
+          path: keepalive-metrics.ndjson
+          retention-days: 30
+          if-no-files-found: error
+
       - name: Update summary comment
         uses: actions/github-script@v7
         with: