[audit-workflows] π Agentic Workflow Audit β 2026-07-01 #42840
Closed
Replies: 1 comment
-
|
This discussion has been marked as outdated by Agentic Workflow Audit Agent. A newer discussion is available at Discussion #43043. |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Daily audit of agentic workflow runs (evening window; log download capped at the ~60s bridge timeout, so 84 of 196 run directories had complete
run_summary.jsonand were analyzed).Overview
Engine Health
Failure Clusters (12 failures β 3 root-cause groups)
All 12 failures were 0-turn / 0-token β i.e. the agent never produced output; failures occurred in setup/CLI-launch or post-agent steps, not during reasoning.
Cluster A β copilot "Execute GitHub Copilot CLI" 0-turn (8 fails) β CHRONIC
Agent job fails at the
Execute GitHub Copilot CLIstep after a few minutes withTurns=0 / Tokens=0 / ErrorCount=0. Maps to known issuescopilot-sdk-driver-failures(recurrence now 20) andsmoke-ci-copilot-cli-100pct-fail-on-push(recurrence now 2 β 2nd consecutive 100%-fail window).Cluster B β NEW: cross-engine "Install Playwright CLI" failures (3 fails)
All three failed at the
Install Playwright CLIagent-setup step in the same window, across three different engines. The cross-engine signature strongly implies an infra/network cause (Playwright browser download / npm registry / firewall allowlist), not an engine bug. New known issue:playwright-install-cli-failure.Cluster C β codex "Process Safe Outputs" failure (1 fail)
Changeset Generator (codex): agent job succeeded, but the
safe_outputsjob failed at theProcess Safe Outputsstep. Consistent with the chronicsafe-output-partial-failure-intolerancepattern (recurrence now 8) where a single safe-output item reds the whole job.π 30-Day Trends
Workflow Health β success/failure counts with success-rate overlay (dashed line = 85% baseline). Volume swings reflect the audit's variable download window; today's 85.7% sits right on the long-run baseline, with the failure share dominated by the copilot Execute-CLI cluster.
Token Usage β daily tokens (M) + 7-day moving average. The shaded region marks the ongoing gap: the token artifact has been empty since 2026-06-19, so per-run token accounting is unavailable and AIC has become the effective usage metric (5,703.6 today). Restoring token reporting would re-enable this chart.
π― Recommended Actions
Execute GitHub Copilot CLI(0-turn). If Smoke CI is the copilot-engine canary, its constant failure onpushlikely signals a real copilot CLI startup regression on main. Add a fast-fail diagnostic that surfaces why the CLI aborts 0-turn.Process Safe Outputsstep continues to red otherwise-green runs (Changeset Generator today). Make per-item safe-output failures non-fatal to the job.β Positives
Repo memory updated
audit-history.jsonl,known-issues.json(+1 new:playwright-install-cli-failure; bumped copilot-sdk, smoke-ci, safe-output recurrences),anomalies.json,recommendations.json,metrics-summary.json,workflow-trends.json. Files compacted to fit the 60 KB memory budget (50.8 KB total, push validated).References:
Warning
Firewall blocked 1 domain
The following domain was blocked by the firewall during workflow execution:
awmgmcpgSee Network Configuration for more information.
Beta Was this translation helpful? Give feedback.
All reactions