[detection-analysis] Detection Analysis Report — 2026-08-07

[github-actions[bot]]

Summary

• Window evaluated: 2026-08-06T12:10:50Z → 2026-08-07T00:44:41Z UTC (~12.6h of the requested last-24h window — see data-coverage note below)
• Total runs analyzed: 259 (212 classified via aw_info.json, 47 with no retrievable metadata)
• Detection-enabled runs: 190 (89.6% of classified)
• Regular runs: 22 (10.4% of classified)
• Misconfigured workflows found: 3

Warning

3 misconfigured workflows found — see table below.

Comparison Chart

Detection-enabled runs had a materially higher success rate (84.7% vs 63.6%) and higher average token usage (45,455 vs 28,323) than regular runs in this window — consistent with detection runs skewing toward the higher-volume, more mature reviewer workflows (Matt Pocock Skills Reviewer, Test Quality Sentinel, PR Code Quality Reviewer, etc.).

Misconfigured Workflows

Workflow	Misconfiguration Type	Run Count	Recommended Fix
Deployment Incident Monitor	Name contains "monitor" but lacks gh-aw-detection: true	2	Add gh-aw-detection: true to frontmatter, or document an explicit repo-level opt-out like the AIC Usage Audit exception
Deep Report	Name contains "report" but lacks gh-aw-detection: true	1	Add gh-aw-detection: true to frontmatter
Daily Regulatory Report Generator	Name contains "report" but lacks gh-aw-detection: true	1	Add gh-aw-detection: true to frontmatter

Daily Agentic Workflow AIC Usage Audit also matches the name-based rule ("audit") but is excluded per its documented repository-level opt-out (source-managed from githubnext/agentic-ops, mirrored without a local override).

No violations were found for:

• Rule 1 (gh-aw-detection: false with >3 runs/7d) — no workflow with detection disabled showed more than 2 runs even in this sampled window; full 7-day data wasn't fetched (cost-prohibitive), so this rule could not be fully evaluated, but nothing in-window suggests a violation.
• Rule 3 (detection enabled but detection steps failed) — 29 detection-enabled runs failed, all classified generically as driver_exit (engine process exit), a reliability issue distinct from the detection feature itself. No evidence tied any failure specifically to a detection step.
• Rule 4 (alternates on/off within the window) — every workflow name observed had a single, consistent gh-aw-detection value across all its runs in the window.

View All Run Metrics

Workflow	Detection	Runs	Success	Failure	Success Rate	Avg Tokens
Matt Pocock Skills Reviewer	Yes	28	28	0	100.0%	15,019
Test Quality Sentinel	Yes	26	25	0	96.2%	13,025
PR Code Quality Reviewer	Yes	25	15	10	60.0%	38,735
Design Decision Gate 🏗️	Yes	24	21	3	87.5%	11,679
Impeccable Skills Reviewer	Yes	23	23	0	100.0%	8,128
PR Sous Chef	Yes	9	7	2	77.8%	37,347
PR Description Updater	Yes	8	8	0	100.0%	14,292
Issue Monster	Yes	6	4	2	66.7%	26,984
Auto-Triage Issues	Yes	4	2	2	50.0%	16,084
AI Moderator	Yes	3	3	0	100.0%	63,652
Avenger	Yes	3	3	0	100.0%	10,558
Contribution Check	Yes	3	2	1	66.7%	28,750
Code Scanning Fixer	Yes	2	1	1	50.0%	8,960
Deployment Incident Monitor	No	2	2	0	100.0%	10,547
[aw] Failure Investigator (6h)	Yes	2	1	1	50.0%	45,943
Discussion Task Miner - Code Quality Improvement Agent	No	2	1	1	50.0%	19,434
PR Triage Agent	No	2	1	1	50.0%	14,978
Daily Regulatory Report Generator	No	1	1	0	100.0%	21,578
Deep Report	No	1	0	1	0.0%	38,163
Daily Agentic Workflow AIC Usage Audit	No (documented opt-out)	1	1	0	100.0%	18,957
~40 other single-run workflows	Mixed	1 each	—	—	—	—

Full per-run raw metrics are available in the pre-downloaded logs under /tmp/gh-aw/aw-mcp/logs/ for this session.

View Historical Trend

No trend chart yet — this is the first recorded data point in the trending history (/tmp/gh-aw/cache-memory/trending/detection-metrics/history.jsonl). At least 7 daily entries are needed before a 30-day trend chart is generated; check back after the trending store accumulates more runs.

Recommendations

1. Add gh-aw-detection: true to the frontmatter of Deployment Incident Monitor, Deep Report, and Daily Regulatory Report Generator, or explicitly document an opt-out if detection is intentionally skipped for these.
2. Investigate the elevated driver_exit failure rate on PR Code Quality Reviewer (10 of 25 runs failed) — this is a reliability issue independent of the detection misconfiguration findings and is the single largest contributor to detection-run failures in this window.
3. Consider widening the pre-downloaded log window in future runs (or increasing the fetch count/pagination budget) — the available data only covered ~12.6 of the requested 24 hours due to the repository's high run volume; roughly 47 runs in that window had no retrievable aw_info.json and were excluded from classification.

References:

• §31132973257 — PR Code Quality Reviewer failure
• §31115328668 — Deep Report failure
• §31131042488 — Daily Regulatory Report Generator

Generated by 🔎 Detection Analysis Report · agent · 164.8 AIC · ⊞ 7.6K · ◷

• expires on Aug 9, 2026, 5:06 PM UTC-08:00

[github-actions[bot]]

This discussion has been marked as outdated by Detection Analysis Report.

A newer discussion is available at Discussion #51234.