[detection-analysis] Detection Analysis Report — 2026-10-29

[github-actions[bot]]

Summary

• Window evaluated: last 24 full hours ending at workflow start (UTC), target range 2026-10-28T23:22:00Z → 2026-10-29T23:22:00Z
• Total runs analyzed (full detail): 453, across 137 distinct workflows
• Detection-enabled runs: 452 (99.8%)
• Regular runs (gh-aw-detection: false): 1 (0.2%)
• Misconfigured workflows found (hard criteria met): 0

Note

No workflow met the hard misconfiguration thresholds in this window. Coverage is partial — see the caveat below before treating this as a clean bill of health.

Coverage caveat: Log collection hit the storage/count budget twice while downloading this window. The 453 runs above have full per-job detail (engine, detection-flag, job conclusions) and span 2026-10-29T03:42Z–23:14Z (~19.5h). An additional 150 runs (mostly from high-frequency workflows like "AI Moderator", "Daily Trajectory Grader Implementer", and "[aw] Failure Investigator (6h)") were only indexed at the GitHub-API metadata level — no aw_info.json was downloaded for them, so their gh-aw-detection status is unknown and they are excluded from the detection classification below. The earliest slice of the 24h window (2026-10-28T23:22Z–~03:42Z) is under-represented. Treat this report as covering the analyzed subset, not the full 24-hour window.

Comparison Chart

Metric	Regular Runs	Detection Runs
Run count	1	452
Success rate	100.0%	79.2%
Avg tokens	11,909	129,211.3
Failures	0	88

The "Regular" group has only a single sample this window, so its 100% success rate is not statistically meaningful — it reflects one run, not a trend.

Misconfigured Workflows

No misconfigured workflows detected in this window against the three hard criteria (explicit opt-out with >3 runs/7d, detection-job-specific failures, or alternating config). Detection-job outcomes among the 452 detection-enabled runs: 423 succeeded, 10 skipped, 0 failed — so rule 2 (detection-step errors) had zero hits.

One item worth watching, below the confirmed-misconfiguration bar:

Workflow	Observation	Run count (window)	Recommended action
Daily GitHub Docs SEO Optimizer	Explicit gh-aw-detection: false, daily schedule	1 (this window)	This window only captured one run, so the >3-runs/7-day threshold can't be confirmed directly — but a daily cron would plausibly cross it over a week. Worth checking over a longer window (§40819484844) whether the opt-out is still intentional.

View All Run Metrics

Top workflows by run count in the analyzed window (full-detail runs only):

Workflow	Runs	Detection-enabled	Regular	Success %	Avg tokens
PR Sous Chef	92	92	0	34.8	5,066
Test Quality Sentinel	31	31	0	100.0	8,848
Issue Monster	30	30	0	86.7	75,462
Ponytail Reviewer	28	28	0	100.0	148,388
PR Code Quality Reviewer	28	28	0	89.3	12,108
Matt Pocock Skills Reviewer	27	27	0	96.3	42,997
Design Decision Gate 🏗️	27	27	0	100.0	1,422
Avenger	19	19	0	100.0	306,924
AI Moderator	13	13	0	92.3	961
Daily Go Test Parallelizer	10	10	0	100.0	1,298,329
Auto-Triage Issues	9	9	0	100.0	108,929
Daily GitHub Docs SEO Optimizer	1	0	1	100.0	11,909

Engine mix across all 453 full-detail runs: copilot (173), pi (172), codex (72), claude (35), aider (1).

View Historical Trend

Detection-enabled run volume has grown from ~175/day on 10-07 to ~450/day currently, while the detection success rate has stayed in the 75–95% band throughout — today's 79.2% is within normal range, not a regression. 19 of the last 30 days have a recorded data point.

Recommendations

1. No action required for detection configuration — the feature is effectively universal (99.8% of classified runs) and no detection-job-specific failures were observed.
2. Revisit "Daily GitHub Docs SEO Optimizer"'s gh-aw-detection: false over a longer window to confirm whether the opt-out is still intentional, since it only showed once in this ~20h slice.
3. Not a detection issue, but worth flagging separately: "PR Sous Chef" had a 34.8% success rate over 92 runs this window (e.g. §40859228926) — this is an agent-logic reliability problem unrelated to the detection feature, but it dominates the failure count in this dataset and would be worth a dedicated investigation.
4. Re-run this analysis with a longer collection budget (or incrementally via continuation) to get full 24h/7-day coverage before treating the "0 misconfigured" result as conclusive — 150 runs in this window remain metadata-only and unclassified.

References:

• §40971672951 (this analysis run)
• §40819484844 (Daily GitHub Docs SEO Optimizer, detection disabled)
• §40859228926 (PR Sous Chef failure example)

Generated by 🔎 Detection Analysis Report · claude · agent · ~141 AIC · ⌖ ~1 AIC · ⊞ 4.4K · ◷

• expires on Nov 1, 2026, 3:33 PM UTC-08:00