Agentic Workflow Audit — 2026-03-24

[github-actions[bot]]

Daily audit of agentic workflow runs for the 24-hour period ending 2026-03-24 21:04 UTC.

Summary

Metric	Value
Total Runs (w/ summaries)	32 of 317 directories
✅ Success	26 (81.25%)
❌ Failure	2 (6.25%)
⚠️ Cancelled	2 (6.25%)
Active Workflows	19
Total Tokens	21,372,711
Estimated Cost	$2.89
Total Turns	329
Missing Tool Reports	1
MCP Failures	0

Workflow Health Trends

Today's audit shows an 81.25% success rate across 32 tracked runs. The success rate is slightly below the typical target range; two workflows experienced failures due to GitHub API errors and an indexing job misconfiguration. No MCP server failures were detected, which is a positive indicator of infrastructure health.

Token Usage & Cost Trends

Token usage is heavily skewed by two outlier runs — CI Cleaner (13.6M tokens, 145 turns) and Sergo - Serena Go Expert (4.4M tokens, 89 turns) — which together account for ~84% of total daily token spend. These two runs warrant attention as potential optimization targets.

---

❌ Failed Runs

Daily Documentation Updater — [§23510638836](https://github.com/github/gh-aw/actions/runs/23510638836) (scheduled, 20:26 UTC)

• Duration: 2.9 minutes
• Failing Job: indexing (2.5 min)
• Error: activation job set outputs (lockdown_check_failed, model, secret_verification_result) but the indexing job subsequently failed, causing all downstream jobs (agent, safe_outputs, update_cache_memory, conclusion) to be skipped
• Root Cause: Likely a model configuration or secret verification issue during the indexing phase
• Recommendation: Investigate the lockdown_check_failed output value; check if the model or secrets for this workflow are correctly configured

Smoke Copilot — [§23509681603](https://github.com/github/gh-aw/actions/runs/23509681603) (PR #22738, branch: copilot/move-apm-packaging-into-own-job)

• Duration: 8.1 minutes
• Agent: Completed successfully (5.2 min)
• Failing Jobs: update_cache_memory + safe_outputs
• Errors:
  1. update_cache_memory: Unable to download artifact(s): Artifact not found for name: cache-memory — first-run or new branch with no prior cache artifact
  2. safe_outputs: 4 outputs failed — discussion comment, issue creation, haiku-printer workflow dispatch, and PR review all failed with Unexpected end of JSON input or Server Error
• Root Cause: Transient GitHub API failures; the cache-memory issue is likely a new-branch first-run condition
• Recommendation:
  • Make cache-memory download step conditional (continue-on-error: true or existence check)
  • Review safe_outputs retry logic for transient GitHub API errors

---

⚠️ Cancelled Runs

Run ID	Workflow	Branch	Event
§23507044143	Smoke Codex	main	scheduled
§23507044155	Smoke Claude	main	scheduled

Both smoke tests were cancelled around the same time (18:56 UTC), suggesting a possible runner capacity issue or deliberate cancellation.

---

🔧 Missing Tool Reports

Tool	Workflow	Run	Severity
add_smoked_label	Smoke Codex	§23507109687	Medium

The add_smoked_label safe-output action is not listed in the Smoke Codex workflow's available tools. The workflow attempted to call it but it was rejected. Fix: Add add_smoked_label to the Smoke Codex workflow's safe-output tools configuration, or replace it with the add_labels tool.

---

🔥 Top Token Consumers

Rank	Run ID	Workflow	Tokens	Turns
1	§23505275817	CI Cleaner	13,637,319	145
2	§23510452858	Sergo - Serena Go Expert	4,386,589	89
3	§23508753002	Step Name Alignment	1,004,115	28
4	§23506527450	Auto-Triage Issues	772,691	18
5	§23511864764	Daily Workflow Updater	378,754	10

Note: CI Cleaner consumed 64% of all tokens today with 145 turns — considerably more than typical for a CI cleanup workflow. This warrants investigation into whether it was operating on an unusually large set of stale runs.

---

Workflow Success Rates

View All 19 Workflows

Workflow	Runs	Success	Failure	Tokens
CI Cleaner	1	✅ 1	—	13,637,319
Sergo - Serena Go Expert	1	✅ 1	—	4,386,589
Step Name Alignment	1	✅ 1	—	1,004,115
Auto-Triage Issues	5	✅ 5	—	1,191,543
Daily Workflow Updater	1	✅ 1	—	378,754
PR Triage Agent	1	✅ 1	—	298,862
Agent Container Smoke Test	2	✅ 2	—	324,651
Issue Monster	1	✅ 1	—	139,682
Smoke Claude	3	✅ 2	—	~150,000
Smoke Codex	3	✅ 2	—	~90,000
AI Moderator	3	✅ 3	—	~0
Smoke Copilot	2	1	❌ 1	—
Daily Documentation Updater	1	0	❌ 1	—

---

Recommendations

High Priority:

1. Investigate Daily Documentation Updater indexing failure — the lockdown_check_failed / secret verification failure pattern suggests a configuration drift. Check model and secret settings for this workflow.
2. Improve Smoke Copilot resilience — add continue-on-error: true to the cache-memory download step for new branches; investigate safe_outputs retry behavior for transient GitHub API errors.

Medium Priority:
3. Add add_smoked_label to Smoke Codex safe-output tools — the missing tool prevents proper smoke test labeling.
4. Review CI Cleaner efficiency — 145 turns and 13.6M tokens is atypically high; consider adding turn/token limits or task scoping.
5. Investigate cancelled smoke runs — Smoke Claude and Smoke Codex were cancelled at 18:56 UTC; confirm this was intentional.

Low Priority:
6. Note on data completeness: Only 32 of 317 run directories contained run_summary.json files. The remaining 285 runs have workflow-logs only and may include many short-lived or pre-agent runs. This is expected behavior.

---

References:

• §23510638836 — Daily Documentation Updater (failure)
• §23509681603 — Smoke Copilot (failure)
• §23507109687 — Smoke Codex (missing tool)

AI generated by Agentic Workflow Audit Agent · history

• expires on Mar 25, 2026, 9:10 PM UTC

[github-actions[bot]]

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

A newer discussion is available at Discussion #22984.