[audit-workflows] 🔍 Agentic Workflow Audit — 2026-07-05 (89.8% healthy; all fails copilot 0-turn CLI)

[github-actions[bot]]

🔍 Agentic Workflow Audit — 2026-07-05

Window: partial ~6h (15:08–21:15Z) · 108 runs analyzed (220 additional run dirs had no downloadable artifacts).

Metric	Value
Overall success	97/108 = 89.8%
Prod-main	40/49 = 81.6%
Non-main (PR/dev)	57/59 = 96.6%
Tokens (in+out)	~3.36M
AI credits (AIC)	~6,311
Missing tools / data / MCP failures	0 / 0 / 0

Headline: Healthy-on-baseline. All 11 failures were copilot-engine, 0-turn/0-tok, failing at the same step: agent >> Execute GitHub Copilot CLI. Both failure clusters are known-chronic — zero novel signatures. claude (20/20), pi (15/15) and codex (1/1) were 100% clean.

📊 Trend Charts

Workflow Health (30d)

Success rate sits at 89.8%, just below the 90% baseline line — consistent with the recent 07-04 rebound (89.1%) and well clear of the 07-02/07-03 troughs (60%/68%) driven by the now-resolved pi collapse. Failures remain a thin, stable band dominated entirely by the copilot engine.

Token Usage (30d)

Daily token volume (~3.36M) is back in a normal band after the metrics-pipeline gap that left run.TokenUsage empty on many mid/late-June days (values here are recovered from token_usage_summary). The 7-day moving average is trending down from the mid-June peaks as high-token experimental workflows have stabilized.

🚨 Failure Clusters (both KNOWN-chronic)

1. Smoke CI — 7/7 = 100% fail on push to main (chronic)

• Engine/model: copilot / claude-sonnet-4.6 · Event: push
• Signature: 0-turn/0-tok, ~3 min, fails at agent >> Execute GitHub Copilot CLI
• Known issue: smoke-ci-copilot-cli-100pct-fail-on-push — now recurrence Add workflow: githubnext/agentics/weekly-research #5, deterministic red on main across 06-30 → 07-05.
• Impact: Fires on every push, so it is a constant red signal on main. If Smoke CI is meant to validate the copilot engine, its 100% failure likely reflects a real copilot-CLI startup regression, not a workflow-config issue.
• Evidence: runs 28745387329, 28748090767, 28750398344, 28750766735, 28751025806, 28753930142, 28754146493.

2. copilot SDK long-run 0-turn — 4 runs (chronic)

Agent does real work (10–24 min), then the job fails 0-turn at Execute GitHub Copilot CLI. Known issue copilot-sdk-driver-failures — now recurrence #23.

Workflow	Event	Branch	Dur	AIC	Note
Daily Safe Output Integrator	schedule	main	24m	74	recurring affected wf
PR Code Quality Reviewer	pull_request	copilot/wrap-org-runne...	19m	221	recurring affected wf
Impeccable Skills Reviewer	pull_request	copilot/spdd-daily-spe...	18m	121	new affected wf name
Linter Miner	schedule	main	10m	50	new affected wf name

Two previously-untracked workflows (Impeccable Skills Reviewer, Linter Miner) now show the same long-run 0-turn signature — added to the known-issue's affected list.

📈 Per-Engine Health

Engine	Runs	Fails	Rate
copilot	72	11	84.7%
claude	20	0	100%
pi	15	0	100%
codex	1	0	100%

Every failure this window is attributable to the copilot CLI startup path. The 07-03 pi collapse (PR Sous Chef) remains resolved — pi 100% for a second consecutive window.

✅ Recommendations (evidence-linked, no new ones)

1. Escalate Execute GitHub Copilot CLI 0-turn startup path to engine owners. Both clusters (Smoke CI deterministic + SDK-longrun across 6 workflows) share the exact same failing step. This is now the single dominant prod-main failure family and has been chronic/unfixed for 5+ observed windows. (Evidence: 11/11 fails at this step; known-issues Add workflow: githubnext/agentics/weekly-research #4 count23, codex naming #14 count5.)
2. Add a fast-fail copilot-CLI startup diagnostic that surfaces why the CLI aborts 0-turn, so long-run workflows fail in seconds rather than burning 10–24 min of agent work first. (Evidence: Daily Safe Output Integrator 24m→0-turn, aic 74 wasted.)
3. No action on missing tools/data — 0 across the window; the tool surface is stable.

Next Actions

• Continue watching Smoke CI on main — if it flips green, the copilot-CLI regression is fixed fleet-wide.
• Watch whether Impeccable Skills Reviewer / Linter Miner become sustained members of the SDK-longrun cluster.

References:

• §28745387329 — Smoke CI (chronic)
• §28751369679 — Daily Safe Output Integrator (longrun 0-turn)
• §28747678796 — PR Code Quality Reviewer (longrun 0-turn)

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 · 235.8 AIC · ⌖ 38.6 AIC · ⊞ 7.6K · ◷

• expires on Jul 6, 2026, 1:44 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 #43881.