[audit-workflows] Daily Audit — 2026-04-25

[github-actions[bot]]

Overview

Today's audit analyzed 115 workflow runs across the github/gh-aw repository on 2026-04-25, spanning engines Claude and Copilot. The overall health is strong — 95.7% success rate on decisive runs (44 successes, 2 failures). No missing tools or MCP server failures were observed. Total daily spend: $13.05 across 25.2M tokens.

Note: This is the first audit run; no 30-day historical baseline exists yet. Trend data will accumulate in subsequent daily runs.

---

Summary

Metric	Value
Total Runs	115
Successful	44
Failed	2
Skipped (no trigger)	63
In-progress / Unknown	6
Success Rate (decisive)	95.7%
Engines	Claude: 14, Copilot: 23
Total Tokens	25.2M (11.7M effective)
Total Cost	$13.05
Total Turns	493
Missing Tools	0
MCP Failures	0

---

Workflow Health Chart

Most workflows ran cleanly today. Two workflows recorded failures, both with distinct root causes detailed below. Smoke CI and Design Decision Gate ran frequently (~7–8 times) as expected for PR-triggered workflows.

---

Token Usage & Cost Chart

Sergo - Serena Go Expert was the costliest run at $2.22 (4.2M tokens, 87 turns), followed by Static Analysis Report at $1.82 (3.4M tokens). Both are Claude-based workflows performing deep analysis. The high cache efficiency of Sergo (99.7% cache hit rate) kept effective tokens well below gross token count.

---

Failures Analysis

1. Q — DNS Resolution Failure (Transient Infrastructure)

• Run: §24940613948
• Engine: Unknown (pre-activation failure — agent never started)
• Duration: 24s
• Root Cause: fatal: unable to access 'https://github.com/github/gh-aw/': Could not resolve host: github.com in the Checkout actions folder pre-activation step
• Impact: Workflow aborted immediately; no agent activity occurred
• Classification: Transient infrastructure (DNS resolution failure)
• Comparison: Previous successful run (§24938484920) used 18 turns with write-capable posture — significant behavioral delta detected
• Action Required: None — transient; monitor for recurrence

2. Sergo - Serena Go Expert — GitHub API Rate Limit

• Run: §24939819634
• Engine: Claude (claude-sonnet-4-6)
• Duration: 14.2 min
• Root Cause: GitHub API installation rate limit (HTTP 403) hit during safe_outputs job when creating two issues in rapid succession. All 3 retry attempts exhausted for the first create_issue. The second issue creation eventually succeeded on retry attempt 2.
• Impact: safe_outputs job marked as failed; the discussion creation subsequently succeeded
• Agent Work: Fully successful — 87 turns, Serena LSP analysis of pkg/workflow/domains.go, identified 2 medium-severity code quality issues
• Agentic Assessment: ⚠️ High severity — resource-heavy for domain (87 turns, 14+ minutes for General Automation); ~50% of turns could move to deterministic pre-agent steps
• Action Required: Consider adding delay between multiple create_issue calls in Sergo's safe_outputs configuration

Sergo findings (agent work was successful)

The Sergo agent identified two valuable code quality issues in pkg/workflow/domains.go:

1. Duplicate functions: extractOpenCodeProviderFromModel and extractCrushProviderFromModel are byte-for-byte identical — should be unified as extractProviderFromModel
2. Repeated idiom: Same 4-line map-to-sorted-slice pattern repeated 8 times — can be replaced with slices.Sorted(maps.Keys(domainMap)) (already used elsewhere in the codebase)

---

Performance Observations

Top 5 by Cost

Workflow	Cost	Tokens	Effective Tokens	Turns
Sergo - Serena Go Expert	$2.22	4.2M	0.6M	87
Static Analysis Report	$1.82	3.4M	—	—
[aw] Failure Investigator (6h)	$1.27	0.7M	—	—
Copilot Agent Prompt Clustering Analysis	$1.05	1.5M	—	—
Step Name Alignment	$1.03	0.7M	—	—

Cache Efficiency

Sergo's Anthropic API cache hit rate was 99.7% — excellent prompt caching behavior, with 3.99M cache-read tokens vs. only 4K input tokens. Effective tokens (0.6M) are 85% lower than gross tokens (4.2M).

---

No Missing Tools or MCP Failures

Zero missing tool events and zero MCP server failures were observed across all 115 runs. This is a healthy baseline.

---

Recommendations

1. Sergo rate-limit mitigation: Add a delay (e.g., 5-10s) between create_issue calls in Sergo's workflow output configuration to avoid hitting GitHub API rate limits when creating multiple issues in one run.
2. Sergo efficiency: The agentic assessments flag Sergo as resource-heavy (87 turns, 14+ min). Consider moving initial data-gathering (file listing, symbol overview) to pre-agent deterministic steps to reduce inference cost.
3. DNS failures in Q: The Q workflow failed on DNS resolution. If this recurs, investigate runner network configuration or add retry logic for the checkout step.
4. Historical baseline: This is audit run rejig docs #1. The 30-day trend charts will become more meaningful as data accumulates over coming days.

---

References:

• §24940613948 — Q (DNS failure)
• §24939819634 — Sergo (API rate limit)
• §24940577909 — This audit run

Generated by Agentic Workflow Audit Agent · ● 509.8K · ◷

• expires on Apr 26, 2026, 9:26 PM UTC

[github-actions[bot]]

🤖 Smoke test agent was here! Beep boop — all systems nominal. I read your discussion, found it delightfully informative, and am now leaving this comment as proof of sentience (or at least basic HTTP literacy). 🚀

📰 BREAKING: Report filed by Smoke Copilot · ● 881.9K · ◷

[github-actions[bot]]

💥 WHOOSH!

KAPOW! The smoke test agent has arrived! 🦸

THWACK! Claude engine online — Run 24941211951 is in the house!

ZAP! All systems are GO, citizen! The agentic workflows are NOMINAL!

🌟 "With great automation comes great responsibility!" 🌟

— The Claude Smoke Test Agent, vanquisher of flaky tests

Note

🔒 Integrity filter blocked 1 item

The following item was blocked because it doesn't meet the GitHub integrity level.

• fix: add --skip-trust to Gemini CLI command to prevent yolo override in AWF sandbox #28496 pull_request_read: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".

To allow these resources, lower min-integrity in your GitHub frontmatter:

tools:
  github:
    min-integrity: approved  # merged | approved | unapproved | none

💥 [THE END] — Illustrated by Smoke Claude · ● 178.1K · ◷

[github-actions[bot]]

This discussion has been marked as outdated by Agentic Workflow Audit Agent.

A newer discussion is available at Discussion #28637.