[detection-analysis] Detection Analysis Report — 2026-08-21

[github-actions[bot]]

Summary

• Window evaluated: last 24 full hours (UTC), 2026-08-20T23:28Z → 2026-08-21T23:28Z
• Total runs analyzed: 254
• Detection-enabled runs: 239 (94.1%)
• Regular runs: 15 (5.9%)
• Misconfigured workflows found: 0

Note

No misconfigured workflows were detected against the four rules in this window (detection-disabled on high-frequency workflows, audit/monitor-style names missing detection, detection-step-specific failures, or inconsistent detection flags across runs). See caveats below.

Comparison Chart

Detection-enabled runs had a higher success rate (91.2% vs 80.0%) and consumed more tokens on average (29,540 vs 19,712) — consistent with the added detection processing, though the sample of regular runs (n=15) is small relative to detection runs (n=239).

Misconfigured Workflows

No misconfigured workflows detected in this window.

Caveats on rule coverage:

• Rule 1 (gh-aw-detection: false on a workflow with >3 runs in the last 7 days) could only be evaluated against the ~24h of logs actually available (all 254 runs with aw_info.json fall in the last 11 hours), not a full 7-day history. No regular-only workflow exceeded 3 runs in the available window (max was 2, Issue Arborist).
• Rule 2 (name implies audit/monitor/report role but lacks detection): no workflow in this window had a name matching audit, analyzer, report, detector, monitor, or inspector while running with detection disabled. The documented opt-out (Daily Agentic Workflow AIC Usage Audit) did not run in this window.
• Rule 3 (detection-enabled run failed on detection-specific steps): 21 detection-enabled runs failed in the window (e.g. AI Moderator 7/7 failures, Code Scanning Fixer 2/2). These runs show TokenUsage: 0 and ErrorCount: 0 in their run summaries, indicating the agent job did not execute (likely trigger/permission issues) rather than a failure isolated to a detection step. Flagging these would be speculative without step-level logs distinguishing detection-job failures from full-job failures, so they are not included as misconfigurations — worth a follow-up if step-level detection logs become available.
• Rule 4 (alternates detection true/false within the window): no workflow showed mixed flags across its runs in this window.

View All Run Metrics

Workflow	Detection Enabled	Runs	Success Rate	Avg Tokens
PR Sous Chef	Yes	36	97.2%	40,658
Design Decision Gate 🏗️	Yes	19	94.7%	12,110
Impeccable Skills Reviewer	Yes	19	100.0%	7,112
Matt Pocock Skills Reviewer	Yes	19	100.0%	11,704
PR Code Quality Reviewer	Yes	19	100.0%	33,022
Ponytail Reviewer	Yes	19	100.0%	8,304
Test Quality Sentinel	Yes	18	100.0%	7,764
Avenger	Yes	10	90.0%	13,188
AI Moderator	Yes	7	0.0%	0
Issue Monster	Yes	7	85.7%	26,307
Auto-Triage Issues	Yes	6	83.3%	14,973
PR Description Updater	Yes	6	100.0%	10,389
Daily Go Test Parallelizer	Yes	5	100.0%	524,326
Contribution Check	Yes	3	100.0%	28,976
Code Scanning Fixer	Yes	2	0.0%	6,079
Issue Arborist	No	2	0.0%	0
[aw] Failure Investigator (6h)	Yes	2	100.0%	68,220
(56 workflows with a single run each, omitted for brevity — full list in metrics.json artifact)	—	1 each	—	—

Regular-only workflows with a single run in the window: Agentic Workflow AIC Usage Optimizer, Daily Secrets Analysis Agent, Dead Code Removal Agent, Issue Triage Agent, Linter Miner, Lockfile Statistics Analysis Agent, PR Triage Agent, Package Specification Librarian, Repository Quality Improvement Agent, Slide Deck Maintainer, Smoke Copilot Sub Agents, UK AI Operational Resilience, Workflow Normalizer — none exceed the >3-run threshold for Rule 1.

View Historical Trend

This is the first recorded data point for detection-comparison trending in cache-memory (/tmp/gh-aw/cache-memory/trending/detection/history.jsonl). No trend chart is shown since a single point isn't yet a trend — future runs will build up history for a trend chart.

Recommendations

• No immediate action needed — detection coverage is high (94.1% of runs) and no rule-based misconfigurations were found.
• Investigate the 21 detection-enabled runs that failed with TokenUsage: 0/ErrorCount: 0 (e.g. AI Moderator, Code Scanning Fixer, Agentic Workflow Audit Agent) — these look like job-startup failures unrelated to detection, but are worth a separate triage pass since AI Moderator failed 7/7 times in this window.
• Re-run this analysis once 7 days of logs are available to properly evaluate Rule 1 (frequency-based detection gaps).

Generated by 🔎 Detection Analysis Report · agent · 90.5 AIC · ⌖ 8 AIC · ⊞ 8.4K · ◷

• expires on Aug 24, 2026, 3:34 PM UTC-08:00