[safe-output-health] Safe Output Health Report - 2026-04-20

[github-actions[bot]]

Executive Summary

• Period: Last 24 hours (2026-04-20, ~12:53–13:11 UTC)
• Total Run Directories Analyzed: 78
• Runs with Agent Execution: 10
• Runs Where Safe Output Jobs Executed: 7
• Safe Output Operations Executed: 10
• Safe Output Operations Failed: 0
• Error Clusters Identified: 0

Overall Status: ✅ All clear — 100% success rate on safe output operations.

---

Safe Output Job Statistics

Job Type	Total Executions	Failures	Success Rate
noop	3	0	100%
add_comment	3	0	100%
push_to_pull_request_branch	2	0	100%
submit_pull_request_review	1	0	100%
create_discussion	1	0	100%
Total	10	0	100%

---

Workflow Activity Summary

Active workflows that ran with safe output jobs:

Workflow	Runs	Safe Outputs Used
Smoke CI	2	noop (×2)
Design Decision Gate 🏗️	2	add_comment + push_to_pull_request_branch (×2)
Test Quality Sentinel	2	noop (×1), add_comment + submit_pull_request_review (×1)
Constraint Solving — Problem of the Day	1	create_discussion (×1)

Runs where safe_outputs job was skipped (71 runs): all skipped because needs.agent.result == 'skipped' — expected behavior when the workflow's detection phase filtered out the event or the workflow was not applicable.

---

Error Clusters

None identified. No safe output job failures were detected in the 24-hour period.

---

Root Cause Analysis

Notable Event: Cancelled Smoke CI Run

One run (§24667588345) was reported with 4 errors in the summary. Investigation found:

• Root cause: The workflow was cancelled before the agent job started (activation.if evaluated to false → agent never ran → artifact "agent" never uploaded).
• Effect: The conclusion job's "Download agent output artifact" step failed with Artifact not found for name: agent, producing a ##[error] marker. This is expected behavior for cancelled runs.
• Handling: The conclusion job correctly detected GH_AW_AGENT_CONCLUSION: cancelled and logged: "Agent job did not fail and no assignment/discussion/code-push errors... skipping failure handling".
• Safe outputs: The safe_outputs job was correctly skipped (Result: false) because needs.agent.result == 'skipped'.
• Classification: Not a safe output health issue. The 4 counted errors are artifact-not-found markers in a cancelled run, not failures in safe output processing.

Recurring digest-mismatch: error Pattern

All 7 runs with safe output execution showed digest-mismatch: error in the Download agent output artifact step's action inputs. Investigation confirmed this is a configuration parameter (digest-mismatch: error) telling the download-artifact action to raise an error on digest mismatch — not an error that occurred. No actual digest mismatches were detected in any run.

---

Recommendations

No Immediate Actions Required

All safe output operations completed successfully. No bugs, configuration changes, or process improvements are warranted based on today's audit.

Observations for Future Monitoring

1. Low volume period: Only 10 safe output operations across 7 workflows in 24 hours. This is within normal range for a development-time window (all runs appear to be during a focused work session around 12:53–13:11 UTC).

2. Skipped job rate is high (91%): 71 of 78 run directories had safe_outputs skipped. This is expected — many are sub-runs from workflows like Test Quality Sentinel and Design Decision Gate that fan out for each PR/event trigger, and most filter out early in detection.

3. Cancelled run error inflation: The error_count: 4 on the cancelled Smoke CI run (§24667588345) inflates the aggregate error count in the summary without representing a real failure. Consider whether cancelled runs should contribute to error counts in reporting.

---

Historical Context

This is the first audit entry in the safe-output-health cache memory. No historical baseline available for trend comparison.

Audit data saved to /tmp/gh-aw/cache-memory/safe-output-health/2026-04-20.json for future trend analysis.

---

Metrics and KPIs

• Overall Safe Output Success Rate: 100%
• Most Reliable Job Type: All job types — no failures in any category
• Most Active Job Types: noop and add_comment (3 executions each)
• Cancelled-run false-positive errors: 1 run (4 reported errors, all expected artifacts of cancellation)

---

Next Steps

• Continue monitoring daily — no immediate actions required
• After 7+ days of data, establish baseline error rate thresholds
• Investigate whether cancelled run errors should be excluded from error_count in summary metrics

References:

• §24667588345 — Cancelled Smoke CI run (false-positive errors)
• §24667650258 — Constraint Solving POTD (create_discussion success)
• §24667646912 — Test Quality Sentinel (add_comment + submit_pull_request_review success)

Generated by Safe Output Health Monitor · ● 494.9K · ◷

• expires on Apr 21, 2026, 1:19 PM UTC

[github-actions[bot]]

This discussion has been marked as outdated by Safe Output Health Monitor.

A newer discussion is available at Discussion #27573.