[audit-workflows] 🔍 Agentic Workflow Audit — 2026-07-13: 86% clean baseline, one 16-workflow startup burst

[github-actions[bot]]

🔍 Agentic Workflow Audit — 2026-07-13

Window: ~7h deep-analyzed slice (14:47–21:31Z), 100 fully-analyzed runs. First audit after a 7-day gap (last: 2026-07-06).

Headline: Raw success 64% looks alarming but is an artifact of one infrastructure event. Excluding it, the fleet is at its normal 86.4% baseline. Zero missing-tools, missing-data, MCP, or agent-logic failures fleet-wide — every failure is a known pre/0-turn driver signature.

Metric	Value
Runs (analyzed)	100
Success / Failure	64 / 36
Raw success rate	64.0%
Rate ex-infra-burst	77.1%
Rate ex-burst & ex-Smoke-CI	86.4% (== 07-06 baseline 86.25%)
AIC (cost)	3,353
Action-minutes	851
GitHub API calls	1,401
Missing tools / data / MCP	0 / 0 / 0

---

🚨 The one thing that matters: mass startup-failure burst

16 distinct workflows failed within a 10-second window (14:47:38–48Z) — all with null engine, duration, and AIC, meaning they failed before any agent turn ran (checkout / startup / image-pull / queue / cancel). This single event is the sole reason raw fleet health reads 64% instead of the clean ~86%.

• Not agent-logic, not missing-tools — pre-agent infrastructure.
• Affected a broad, unrelated mix: Daily Cli Tools Tester, Artifacts Usage Report, Skillet, Security Review Agent, Smoke Service Ports, GitHub Remote MCP Authentication Test, Matt Pocock Skills Reviewer, Approach Validator, Terminal Stylist, +7 more.
• Hypothesis: a shared runner/queue event, plausibly related to the recent runner-topology infra change (Propagate runner topology into detection AWF image pre-pull #45274, detection AWF image pre-pull, merged shortly before this window).
• Classification pending: need the next 1–2 audits to tell one-time blip from a startup regression.

All 16 burst workflows + timestamps

All at 2026-07-13T14:47:38–48Z, conclusion=failure, engine/duration/aic all null:
Daily Cli Tools Tester, Artifacts Usage Report, Workflow Portfolio Analyst, Skillet, Security Review Agent 🔒, Smoke Service Ports, Dev, GitHub Remote MCP Authentication Test, Matt Pocock Skills Reviewer, Approach Validator, Copilot PR Conversation NLP Analysis, Daily Compiler Quality Check, Smoke Agent: scoped/approved, Terminal Stylist, Daily AWF Spec Compiler Surfacing Review, Documentation Noob Tester.

---

📉 Chronic & recurring (all known, all 0-turn driver)

• Smoke CI — 9/9 = 100% fail on push/PR (smoke-ci-copilot-cli-100pct-fail-on-push, recur Weekly Research Report: AI Workflow Automation Landscape and Market Opportunities - August 2025 #7). copilot CLI 0-turn ~3m. Unfixed since 06-30 (13+ days). Dominant non-burst failure.
• copilot-sdk longrun-0turn (copilot-sdk-driver-failures, recur Fix invalid YAML because of ANSI escape sequences #26) — leading agentic fail family: PR Code Quality Reviewer ×2, PR Triage Agent (15m/aic181), Daily Project Performance Summary Generator (22m/aic147), Smoke Copilot Sub Agents, Daily Issues Report Generator, Daily Agent of the Day Blog Writer. Real work 6–22m then job fails.
• ⚠️ pi PR Sous Chef re-surfaces (pi-gpt54-0tok-agentjob-fail) — 2 fails (8–9m, 0-turn). Was RECOVERED-WATCH since 07-04. This exact pattern escalated into the 07-03 pi-engine collapse (100%→30.8%). Early-warning watch.
• claude longrun-0turn — Workflow Portfolio Analyst (12.9m/aic114), same cross-engine signature as the copilot longrun family.
• token-usage-reporting-gap (recur Add workflow: githubnext/agentics/weekly-research #4) — no fleet token total surfaced again; AIC available but token telemetry empty.

✅ Engine health

Engine	Runs	Success	Rate
codex	3	3	100%
claude	12	11	91.7%
pi	11	9	81.8%
copilot	57	41	71.9% (85.4% ex-Smoke-CI)
null (burst+CI)	17	0	0%

---

📊 Trends (30 days)

Fleet success rate oscillates in a healthy 82–95% band; today's 64% dip is an outlier driven entirely by the 14:47Z startup burst — the clean baseline (86.4%) sits right on the dashed reference line. Run volume is lower today because this is a partial ~7h analyzed slice, not a full 24h fetch.

Token telemetry is sparse (only 10 of 22 days reported) — the declining 7-day MA reflects the token-usage-reporting-gap known issue, not a real drop in usage; recent "low" days are partial-window artifacts. Today's fleet token total was again unavailable (AIC=3,353 used as the cost proxy).

---

🎯 Recommended actions

1. [HIGH · NEW] Triage the 14:47Z startup burst — pull GH Actions runner/queue logs for that 10s window; verify whether Propagate runner topology into detection AWF image pre-pull #45274 (runner-topology / image pre-pull) caused a batch startup/image-pull failure. Annotate if one-time; escalate if it recurs.
2. [HIGH · day13+] Smoke CI 100%-fail-on-push — fix the copilot-CLI-on-push startup path, or gate/quarantine Smoke CI so it stops dominating the health signal.
3. [HIGH · day30+] copilot-sdk-driver-failures — still the leading agentic fail family, now cross-engine (claude/pi share the longrun-0turn signature). Allowlist routine read-only/branch-setup tool calls + flush emitted safe-output on session-idle instead of reddening.
4. [MED · WATCH] pi PR Sous Chef 0-turn — re-watch next audit; pre-emptively inspect the pi model+driver path before it can repeat the 07-03 collapse.
5. [LOW] token-usage-reporting-gap — persistent; fleet token totals still not surfaced.

Repo-memory updated

audit-history.jsonl (+1), known-issues.json (recur bumps + new mass-startup-failure-burst), anomalies.json (new anom-mass-startup-burst-2026-07-13), recommendations.json (+2), workflow-trends.json, metrics-summary.json. Memory trimmed to 57 KB (limit 60 KB) and pushed successfully.

References: §29285932420 · §29259459269 · §29286408558

Warning

Firewall blocked 1 domain

The following domain was blocked by the firewall during workflow execution:

• awmgmcpg

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "awmgmcpg"

See Network Configuration for more information.

Generated by 🔍 Agentic Workflow Audit Agent · 263 AIC · ⌖ 36.6 AIC · ⊞ 7.4K · ◷

• expires on Jul 14, 2026, 1:47 PM UTC-08:00

[github-actions[bot]]

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

A newer discussion is available at Discussion #45558.