[safe-output-health] 🏥 Safe Output Health Report — 2026-05-13

[github-actions[bot]]

Executive Summary

• Period: Last 24 hours (ending 2026-05-13)
• Runs analyzed: 69
• Engines: copilot (41), claude (11), codex (2), other (15)
• Safe-output jobs failed: 0
• safeoutputs MCP error rate: 0.0%
• Status: ✅ Healthy — no safe-output job failures detected in the audit window

The safe-output subsystem is operating cleanly. All safe_outputs jobs across the sampled runs completed with conclusion=success, including the two runs where the upstream agent job failed. The safeoutputs MCP server reported a 0% error rate across every audited run.

Safe-Output Job Statistics

Aggregate counts of safe-output tool calls observed across episodes in the window. Job-level success rate is 100% for every job audited.

Tool	Calls observed	Status
noop	2+	✅
add_comment	2	✅
push_to_pull_request_branch	1	✅
submit_pull_request_review	2	✅
create_pull_request_review_comment	4	✅
create_issue	1	✅
update_issue	1	✅
missing_data	1	✅
safe_outputs job conclusion	all success	✅

Error Clusters

No safe-output job error clusters were identified in this window.

For context, the 10 errors counted in the workflow summary come from runs whose safe_outputs job itself was not the failure point:

Where the 10 errors came from (out of scope for this monitor)

Run	Workflow	Errors	Conclusion	Notes
§25780203032	Smoke CI	4	cancelled	Regular CI workflow (not agentic) — out of scope
§25778833286	Smoke CI	3	cancelled	Regular CI workflow — out of scope
§25773858514	Smoke CI	1	cancelled	Regular CI workflow — out of scope
§25779502806	Step Name Alignment	1	failure	Agent job failed; safe_outputs job: success
§25772016504	Design Decision Gate 🏗️	1	failure	Agent job failed; safe_outputs job: success

The Smoke CI errors come from the regular CI workflow (test cancellation), not from agentic workflows. The Step Name Alignment and Design Decision Gate failures are agent-job failures handled by separate monitors; in both cases the dedicated safe_outputs job still completed successfully.

Root Cause Analysis

• API errors: None observed. safeoutputs MCP server returned error_count=0 across all audited runs.
• Parsing errors: None observed. agent_output.json files were validated successfully in every audited run.
• Validation errors: None observed.
• Permission errors: None observed (firewall blocked=0 across audited runs).
• Logic errors in safe-output scripts: None observed.

Recommendations

Immediate actions (critical)

None. Safe-output health is green for this window.

Process Improvements

1. Confirm that agent-job failures do not silently swallow safe-output emissions. Both §25779502806 and §25772016504 had a failing agent job but a successful safe_outputs job; the audit description claimed the workflow "failed before agent activation" while turn counts (31, 21) clearly indicate the agent ran. Worth verifying the audit-report wording lines up with actual job graph state, since misleading descriptions undermine triage by other monitors. Severity: low (cosmetic, not affecting safe-output execution).
2. Track the small absolute volume of safe-output tool calls. Across 69 runs, only ~13 safe-output tool invocations were observed (plus several noop calls). This is consistent with a read-only/analysis-heavy workload, but worth keeping in the history so trends are visible if a regression silently suppresses emissions.

Work Item Plans

No work items required for this window. A monitoring baseline has been recorded in /tmp/gh-aw/cache-memory/safe-output-health/ for trend comparison in future audits.

Historical Context

This is the first audit stored in the safe-output-health cache memory. Future audits should diff against 2026-05-13.json for trend analysis: error rate change, new tool types appearing, job conclusion regressions.

Metrics

• Overall safe-output job success rate: 100% (all audited jobs)
• safeoutputs MCP error rate: 0.0%
• Most-used safe-output tool: create_pull_request_review_comment (4 calls in single run §25772016551)
• Most diverse safe-output episode: §25772298094 — [aw] Failure Investigator used create_issue + update_issue

Next Steps

• Re-run audit in 24h and compare against today's baseline
• Flag any new tool types that appear (potential schema changes)
• Continue to scope this monitor strictly to safe-output jobs (agent-job and detection-job failures handled by other workflows)

References:

• Audit run: §25780240323
• Sample heavy safe-output use: §25772016551
• Sample agent-fail/safe-output-ok: §25779502806

Generated by Safe Output Health Monitor · ● 25.4M · ◷

• expires on May 14, 2026, 5:38 AM UTC

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-05-14T05:38:18.327Z.

Closed by Workflow