[detection-analysis] Detection Analysis Report — 2026-09-11 #60352
Closed
Replies: 1 comment
|
This discussion has been marked as outdated by Detection Analysis Report. A newer discussion is available at Discussion #60483. |
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.
Summary
2026-09-10T23:36Z→2026-09-11T23:36Zgh-aw-detection: false): 12 (2.7%)Note
No misconfigured
gh-aw-detectionworkflows were found in this window. All 12 explicit opt-outs are single-run smoke-test workflows well under the >3-runs threshold for flagging, and nodetectionjob showed a failing conclusion.Comparison Chart
Regular runs had a 50.0% success rate vs. 51.2% for detection-enabled runs — no meaningful difference in reliability between the two groups this window. Average token usage reported as 0 for both groups — see caveat below.
Misconfigured Workflows
No misconfigured workflows detected in this window.
View All Run Metrics
Top workflows by run count (177 distinct workflows total in window):
Explicit
gh-aw-detection: falseopt-outs seen this window (all single-run, below the flagging threshold): Agent Container Smoke Test, Smoke Pi, Smoke Codex, Smoke Copilot Auto, Smoke Claude, Smoke Copilot, Smoke Copilot - AOAI (apikey), Smoke Copilot - AOAI (Entra), Smoke Copilot MAI, Smoke Copilot Small, Smoke OTEL, Daily GitHub Docs SEO Optimizer.Data caveat: 165 additional runs in the window (mostly non-agentic CI jobs and a few agentic runs, e.g.
AI Moderatordependabot triggers) had only a rawrun.json— noaw_info.json/run_summary.json— so they could not be classified by detection status and are excluded from the totals above.Token usage caveat:
TokenUsagereads as0for every one of the 442 runs in this window, for both groups. Historical trend entries from prior days show non-zero averages (tens of thousands to ~250K tokens/run), so this looks like a telemetry gap in today's log batch rather than a real drop in usage — worth a follow-up check on the metrics pipeline rather than treating it as a detection-feature effect.View Historical Trend
Detection-run volume has grown from ~250/day to ~430/day over the last ~2.5 weeks, while regular (opted-out) runs have stayed near-zero throughout. Detection success rate has ranged roughly 60–93% day to day.
Recommendations
gh-aw-detectionfeature is being used as intended (default-on, with a small, appropriately-scoped set of smoke-test opt-outs).TokenUsageis reporting0across all 442 runs today — compare against the log-collection/metrics pipeline that populatesrun_summary.json'srun.TokenUsagefield, since prior days show real non-zero values.aw_info.json/run_summary.json(mostly non-lock-file CI jobs, but including at least one agentic run) can be classified in future reports.References:
gh-aw-detection: falseopt-outWarning
Firewall blocked 1 domain
The following domain was blocked by the firewall during workflow execution:
api.anthropic.comTo allow these domains, add them to the
network.allowedlist in your workflow frontmatter:See Network Configuration for more information.
All reactions