[safe-output-health] 🏥 Safe Output Health Report - 2026-08-05

[github-actions[bot]]

Executive Summary

• Period: Last 24 hours (pre-downloaded log set covering runs from 2026-08-05T04:14Z to 2026-08-05T05:08Z)
• Runs Analyzed: 18
• Workflows Active: 16 distinct workflows (Sergo - Serena Go Expert, jsweep - JavaScript Unbloater, Smoke Kiro x3, Step Name Alignment, PR Sous Chef, Designer Drift Audit, Copilot CLI Deep Research Agent, AI Moderator, Documentation Noob Tester, Smoke Cursor x2, Refactoring Cadence, Code Simplifier, Go Logger Enhancement, Issue Monster, PR Description Updater)
• Safe Output Jobs Executed: 18 (every run has exactly one unified safe_outputs job)
• Safe Output Jobs Failed: 0
• Error Clusters Identified: 0

Bottom line: clean bill of health. Every safe_outputs job in the last 24 hours completed successfully -- no create_issue, create_discussion, add_comment, update_pull_request, or push_to_pull_request_branch failures were observed. This is a marked improvement from the 2026-08-04 audit, which found 2 safe_outputs job failures (partial-batch failures in PR Sous Chef and Design Decision Gate) across a much larger 200-run sample.

Safe Output Job Statistics

Job Type	Total Items Processed	Failures	Success Rate
create_issue	4	0	100%
add_comment	4	0	100%
update_pull_request	2	0	100%
create_discussion	1	0	100%
safe_outputs job (all types, unified)	13 items / 18 job runs	0	100%

gh-aw runs a single unified safe_outputs job per workflow run -- one Process Safe Outputs step batches every configured safe-output type for that run rather than dispatching separate jobs per type (create_issue, add_comment, etc. are handled as steps/messages within that one job).

Error Clusters

None. No safe_outputs job failures, no Process Safe Outputs step failures, and no partial-batch failures (where some items in a batch succeed while others fail) were found in any of the 18 runs.

Root Cause Analysis

Not applicable this cycle -- there is nothing to root-cause. For reference, the two clusters found in the prior audit (2026-08-04) were:

• Validation error: resolve_pull_request_review_thread rejecting a PullRequestReviewComment node ID where a PullRequestReviewThread node ID was required.
• Validation/config error: push_to_pull_request_branch blocked because the agent patch touched files outside the workflow allowed-files allowlist.

Neither pattern recurred today (0 occurrences in 18 runs).

Recommendations

Critical Issues (Immediate Action Required)

None.

Bug Fixes Required

None identified this cycle.

Configuration Changes

None required.

Process Improvements

1. Partial-batch-failure visibility (carried over from 2026-08-04, still proposed, not yet re-verified since no failures occurred to check against)
   • Current State: The Process Safe Outputs step exits non-zero if any item in a multi-item batch fails, even when most items succeeded and were persisted to GitHub. This makes job-level pass/fail a noisy proxy for whether the safe outputs actually worked.
   • Proposed: Expose a partial-success indicator (distinct step outcome or job summary annotation) instead of a binary failure when items_succeeded > 0.
   • Benefit: Reduces false-alarm red X marks on runs that mostly succeeded; unaffected by today clean run but still worth tracking since it is about how future partial failures get surfaced.

Out-of-Scope Failures Observed (noted for awareness only, not analyzed further)

5 of the 18 runs failed overall, but every failure was in a job other than safe_outputs, so they are out of scope for this audit per its charter (safe-output jobs only):

Run	Workflow	Failing Job	Failing Step
30975982770	Smoke Kiro	agent	Execute Kiro CLI
30974811087	Smoke Kiro	agent	Execute Kiro CLI
30974539538	Smoke Cursor	agent	Execute Cursor Agent CLI
30977137021	Smoke Kiro	check_token_telemetry	Assert token_usage.jsonl is non-empty
30975573886	Smoke Cursor	check_token_telemetry	Assert token_usage.jsonl is non-empty

In every one of these 5 runs, the safe_outputs job itself still ran and completed with conclusion: success (it simply had nothing to process, or processed successfully, independent of the upstream agent/telemetry-check failure). These are agent-driver-exit and telemetry-guardrail issues, tracked by other monitoring, not safe-output job health.

Historical Context

Comparing against the 2026-08-04 audit (200 runs analyzed, 2 safe_outputs failures):

Trends

• Error rate trend: Improving -- 2/200 (~1%) failing safe_outputs jobs on 2026-08-04 to 0/18 (0%) today. Sample sizes differ substantially (200 vs 18 runs) since this audit worked from a smaller pre-downloaded log window, so this is directional, not a rigorous rate comparison.
• Most common recurring issue: None recurring -- both 2026-08-04 patterns (resolve-thread-wrong-node-id, push-branch-outside-allowed-files) are still single-occurrence and did not reappear today.
• Improvement since last audit: Both known patterns show 0 new occurrences.

Metrics and KPIs

• Overall Safe Output Job Success Rate: 100% (18/18)
• Most Reliable Job Type: All types tied at 100% (create_issue, add_comment, update_pull_request, create_discussion)
• Most Problematic Job Type: None
• Average Time to Failure: N/A (no failures)

Next Steps

• No immediate action required -- safe output pipeline is healthy this cycle.
• Keep resolve-thread-wrong-node-id and push-branch-outside-allowed-files on the watch list; promote to recurring if either reappears in a future audit.
• Consider revisiting the partial-batch-failure visibility proposal from 2026-08-04 the next time a partial failure actually occurs, to validate the recommendation against a live example.
• The 5 non-safe-output failures above (Smoke Kiro / Smoke Cursor agent-driver exits, check_token_telemetry guardrail) are flagged for the agent/detection-job monitoring workflow, not for follow-up here.

Generated by 🔒 Safe Output Health Monitor · agent · 226.5 AIC · ⌖ 33.4 AIC · ⊞ 6.6K · ◷

• expires on Aug 5, 2026, 9:29 PM UTC-08:00

[github-actions[bot]]

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

A newer discussion is available at Discussion #50767.