[audit-workflows] Daily Audit — 2026-04-13

[github-actions[bot]]

Overview

Audit of 115 workflow runs from 2026-04-13 (today's batch). Overall health is 80% success rate (32 success / 4 failure, excluding 75 skipped and 4 in-progress). The dominant root cause of all 4 failures was GitHub API rate limiting caused by concurrent workflow execution. No missing tools or missing data were reported. Token consumption was concentrated in one run: Daily Workflow Updater (459K tokens, $4.56 estimated cost).

Note on trend charts: matplotlib is not available in this runner environment. Charts will be added in a future audit once the dependency is resolved. Audit data has been seeded to repo memory for trend tracking going forward.

---

Health Summary

Metric	Value
Total Runs	115
✅ Success	32 (80%)
❌ Failure	4 (20%)
⏭️ Skipped	75
🔄 In Progress	4
Total Tokens	12,478,749
Estimated Cost	$4.56
Missing Tools	0
Missing Data	0
Total Errors	0 (reported)

---

Failures (4 runs)

All 4 failures share the same root cause: GitHub API rate limit exceeded for installation — triggered by many workflows firing in overlapping windows around 20:30–20:35 UTC.

Run	Workflow	Root Cause
§24364460297	Copilot Agent Prompt Clustering Analysis	Rate limit exceeded accessing artifacts + script exit code 1
§24364571228	Daily DIFC Integrity-Filtered Events Analyzer	Rate limit exceeded downloading artifacts; tried to upload non-existent chart PNG
§24365402913	AI Moderator	Rate limit exceeded during PR checkout for issue_comment trigger
§24365515384	AI Moderator	Rate limit exceeded during issue lock for issue #26074

Recommendations:

• Stagger daily/scheduled workflow cron times to avoid concurrent API bursts (e.g., offset by 5–10 minutes)
• Add retry logic with exponential backoff in activation steps that call GitHub API
• For Daily DIFC Integrity-Filtered Events Analyzer: add a file-existence guard before calling upload_artifact to prevent failures when chart generation is skipped

Rate Limit Error Detail (AI Moderator)

##[error]Failed to lock issue: API rate limit exceeded for installation.
##[error]ERR_NOT_FOUND: Failed to lock issue #26074: API rate limit exceeded for installation.
##[error]❌ Rate limit check failed: API rate limit exceeded for installation.

Timestamp: 2026-04-13T20:35:00–16 UTC — multiple workflows overlapping in a ~16-second window.

---

Medium-Severity Agentic Issues

Daily Workflow Updater — §24366711283

Agent: Copilot | Status: ✅ Success | Duration: 5m43s | Turns: 12 | Tokens: 459K

Assessment	Severity	Evidence
resource_heavy_for_domain	MEDIUM	12 turns, 5m43s for a repo maintenance task
poor_agentic_control	MEDIUM	Exploratory execution style, friction=0
partially_reducible	LOW	~50% turns are data-gathering (agentic_fraction=0.50)
model_downgrade_available	LOW	Repo maintenance domain; frontier model may not be needed

Recommendations:

1. Move data-fetching steps to pre-agent frontmatter steps writing to /tmp/gh-aw/agent/ — this would cut turns by ~50%
2. Consider engine.model: gpt-4.1-mini or claude-haiku-4-5 to reduce inference cost
3. Add tighter stopping conditions / tool restrictions to reduce exploratory turns

---

Workflow Coverage

All Active Workflows (21 distinct)

Workflow	Runs	Status
AI Moderator	3	1 success, 2 failure
Daily Workflow Updater	1	success
Issue Monster	1	success
Auto-Triage Issues	1	success
Copilot Agent Prompt Clustering Analysis	1	failure
Daily DIFC Integrity-Filtered Events Analyzer	1	failure
Grumpy Code Reviewer 🔥	5	skipped
Q	3	skipped
Archie	3	skipped
Scout	3	skipped
Resource Summarizer Agent	3	skipped
/cloclo	3	skipped
Test Quality Sentinel	3	skipped
Design Decision Gate 🏗️	3	skipped
Documentation Unbloat	2	skipped
Contribution Check	1	in_progress
Agentic Workflow Audit Agent	1	in_progress
Copilot CLI Deep Research Agent	1	in_progress
Poem Bot	1	skipped
Mergefest	1	skipped
Plan Command	2	skipped

Overkill Assessments (Skipped Runs)

23 skipped runs received overkill_for_agentic LOW assessments. These are primarily PR/issue comment triggers where the event didn't match the workflow's filter conditions — resulting in zero turns, zero tool calls, read-only actuation. This is expected behavior; the assessment is informational only and does not indicate a problem.

---

Patterns (First Baseline)

This is the first audit run — repo memory has been seeded with today's data. Future runs will surface trends and regressions. Key patterns to watch:

1. API rate limiting — present today across 4 runs; will track frequency
2. Daily Workflow Updater efficiency — 12 turns for repo maintenance is a regression target

---

References:

• §24366943202 (this audit run)
• §24364571228 (DIFC failure)
• §24366711283 (Daily Workflow Updater)

Generated by Agentic Workflow Audit Agent · ● 376K · ◷

• expires on Apr 14, 2026, 9:18 PM UTC