[audit-workflows] Agentic Workflow Audit — 2026-11-11 (partial coverage)

[github-actions[bot]]

Summary

The requested 24-hour log collection returned a partial 100-run sample spanning 2h 57m. Of 90 executed runs, 63 succeeded: 70.0% raw success. The intentional_failure marker was absent, so the adjusted rate cannot be calculated; neither named credit-guardrail workflow appeared by name in this sample. The 30-day charts could not be completed from the available logs.

Key metrics and findings

• Engines: Copilot 53, Pi 24, Codex 13, Claude 4, Aider 1. Counts come from the logs summary derived from engine_id.
• Failures: 29 errors, including 19 driver-exit failures and 4 agent-logic failures; 0 missing tools, 0 missing data, and 0 warnings.
• Tokens: 5,324,873 across the sampled runs. The largest observed run used 751,225 tokens in Daily Go Test Parallelizer.
• Recurring reliability issues: 13 failures in the Handle agent failure conclusion step; 3 driver failures in Auditor evidence collection; 5 workflows failed at shared MCP Gateway startup. Safe-output upload/processing failures also appeared.
• Collection limit: the initial response covered 2026-11-11 18:20:30–21:17:23 UTC. Continuation failed because run 43021931080 exceeded the 64 MiB ATIF input limit. This repeats the partial-coverage finding from recent audits.

Trend charts

The available sample shows a 70.0% raw success rate (63 successes and 27 failures among 90 executed runs) for one partial day. It does not represent a complete 24-hour or 30-day trend; the intentional-failure marker was missing.

The sample totals 5.32M tokens over the same partial window, with a 751k-token run from Daily Go Test Parallelizer. A 7-day moving average is unavailable because only one partial day was collected.

View full findings

• Failure cluster failure-e121990520ae929c groups 13 failures at the Handle agent failure conclusion step across multiple workflows.
• Failure cluster failure-5ea31635c811b70b groups 5 failures at Start MCP Gateway, also across multiple workflows.
• Failure cluster failure-a7fc54d02b71f423 records 3 Auditor evidence-collection failures.
• Failure cluster failure-5df4f6d9ddb75c0d records a Matt Pocock Skills Reviewer safe-output upload failure; failure-f3527e2bef0656a6 records a PR Code Quality Reviewer safe-output processing failure.
• Only 6 of 11 failure clusters had evidence coverage. The logs returned collection_truncated, missing_recorded_parent, and missing_recorded_root coverage reasons.

Recommended actions

1. Repair log pagination and isolate or safely bound oversized sessions so one export cannot stop the remaining 24-hour and 30-day collection.
2. Triage MCP Gateway startup across the five affected workflows and the shared Auditor evidence-collection step.
3. Review the Handle agent failure conclusion behavior and the safe-output failures.
4. Review Daily Go Test Parallelizer token use against its normal run profile.

References: §43099392497 · §43089863483 · §43099282403

Generated by 🔍 Agentic Workflow Audit Agent · codex · gpt60 · ~4 AIC · ⌖ ~1 AIC · ⊞ 16.6K · ◷

• expires on Nov 12, 2026, 1:33 PM UTC-08:00

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-11-12T21:33:11.796Z.

Closed by Workflow