[audit-workflows] Agentic Workflow Audit — 2026-10-21

[github-actions[bot]]

Summary

The 200-run sample recorded an 82.54% raw success rate and 82.45% after excluding one intentional-failure run. Driver exits were the main failure signal, while no missing tools were reported. Collection limits left most of the requested 24-hour window and 30-day history unavailable, so these results are partial and do not establish fleet-wide trends.

24-hour audit sample

• Coverage: 200 unique runs from 2026-10-21, sampled 11:44–14:17 and 14:21–21:03 UTC. The request did not reach the full 24-hour window.
• Health: 189 executed runs; 156 successes and 33 failures. Raw success rate was 82.54%; adjusted success rate was 82.45% after excluding one Daily Credit Limit Test run marked as an intentional failure.
• Engine counts: Copilot 91, Pi 47, Codex 38, Claude 11, Aider 2, using the logs summary engine counts.
• Failure signals: 29 driver-exit failures and 4 agent-logic failures across the sample. Driver exits clustered around Design Decision Gate, Matt Pocock Skills Reviewer, Ponytail Reviewer, PR Code Quality Reviewer, and Test Quality Sentinel.
• Tool and volume metrics: 0 missing tools, 60 errors, 0 warnings, and 22,856,178 tokens.

30-day charts

The available daily bucket is October 21, with 155 adjusted successes, 33 executed failures, and an 82.45% adjusted success rate. The other 29 days were not retrieved, so this chart cannot show a 30-day trend.

The retrieved October 21 sample totals 22.86 million tokens. Only one daily bucket is available, so the requested 7-day moving average cannot be calculated and no longer-term token trend can be inferred.

Collection limits and recommendations

The 30-day request stopped at the 100-run count limit after covering about six hours. The initial 24-hour request also returned 100 runs; a continuation added another 100 runs, while a higher-count request timed out after five minutes.

• Investigate driver exits in the five reviewer and gate workflows listed above.
• Improve log collection pagination or timeout behavior, then rerun the 24-hour audit and regenerate the charts from complete 30-day run summaries.
• Preserve the distinction between intended guardrail failures and operational failures in all success-rate rollups.

Generated by 🔍 Agentic Workflow Audit Agent · codex · gpt60 · 3.63 AIC · ⌖ 0.352 AIC · ⊞ 16K · ◷

• expires on Oct 22, 2026, 1:40 PM UTC-08:00

[github-actions[bot]]

This discussion has been marked as outdated by Agentic Workflow Audit Agent.

A newer discussion is available at Discussion #71226.