[workflow-analysis] Weekly Workflow Analysis — Apr 20–27, 2026

[github-actions[bot]]

Weekly analysis of 43 workflow runs from Apr 20–27, 2026. Overall health is moderate: 67% success rate with two recurring infrastructure patterns driving the majority of failures, and widespread resource-heaviness that inflates cost.

Summary

Metric	Value
Total runs	43
✅ Success	29 (67%)
❌ Failure	8 (19%)
⏹ Cancelled	4 (9%)
🔄 In Progress	2 (5%)
Total cost	$12.09
Total tokens	36.3M (21.2M effective)
Total action-minutes	341
Engine split	copilot: 29 · claude: 7 · codex: 5
Errors logged	22

---

Critical Issues

❌ Pattern 1 — node: command not found in chroot mode (exit code 127)

Affected: Daily News (§24986870660)

The copilot-driver launches inside a chroot container but node is not on the PATH after the chroot pivot. The agent never starts; the run fails immediately with:

/bin/bash: line 1: node: command not found
[WARN] Command completed with exit code: 127

This is a hard infrastructure failure — zero agent turns, zero tokens, no useful output. The same root cause has been observed in other copilot-engine chroot runs. The fix requires ensuring Node.js is resolvable inside the chroot (via GH_AW_NODE_BIN, toolcache symlink, or explicit PATH injection before the chroot pivot).

---

❌ Pattern 2 — Visual Regression Checker docs-build step failure on add-shadow-ops PR

Affected: 4 runs across 2 PR pushes on branch add-shadow-ops (#28676)

• §24986681062 — failure
• §24986645470 — cancelled (superseded)
• §24983845728 — failure
• §24983824636 — cancelled (superseded)

The Build documentation pre-agent step consistently exits non-zero on this branch. The agent is never reached (no tokens, no turns). Two failure+cancel pairs reflect the pattern of the PR receiving a new push while a run is in flight, making the first run redundant.

Resolution: Fix the docs build on the add-shadow-ops branch or scope the Visual Regression Checker trigger to skip the docs-build step when the docs paths haven't changed.

---

❌ Safe Outputs Step Failures (post-agent)

[aw] Failure Investigator (§24982458545) — The agent succeeded (85 turns, issued parent report + sub-issue #28673/#28674, cost $2.95) but the safe_outputs job failed. The agent itself ran correctly.

Schema Feature Coverage Checker (§24981796377) — The agent completed but safeoutputs.create_pull_request was rejected. Cause: protected-files config blocks PRs targeting schema-demo-*.md paths. The Failure Investigator already identified this as P0 and recommended adding allowed-files: [".github/workflows/schema-demo-*.md"] to the workflow's safe-outputs.create-pull-request config. See auto-issue #28671 for details.

---

Performance Issues

Top 10 Token-Consuming Runs

Workflow	Tokens	Effective Tokens	Turns	Duration	Result
Schema Consistency Checker	8.1M	1.2M	138	13.1m	✅
Documentation Unbloat	4.8M	0.79M	58	34.8m	❌
[aw] Failure Investigator	4.7M	0.88M	85	17.9m	❌
Go Fan	2.0M	0.42M	—	—	✅
Agentic Observability Kit	2.0M	2.24M	—	28.6m	✅
Layout Specification Maintainer	1.6M	1.75M	—	—	✅
Organization Health Report	1.3M	1.45M	—	13.8m	✅
Contribution Check	1.3M	1.43M	—	—	✅
Daily MCP Tool Concurrency Analysis	1.2M	1.35M	21	5.3m	🔄
Copilot PR Prompt Pattern Analysis	0.98M	1.10M	—	—	✅

Schema Consistency Checker stands out: 8.1M tokens for 138 turns in 13.1 minutes, with only ~14.9% cache efficiency. This is the most expensive single run this week and warrants a dedicated efficiency review.

Slow Runs (>10 minutes)

Workflow	Duration	Result	Tokens
Documentation Unbloat	34.8m	❌	4.8M
Agentic Observability Kit	28.6m	✅	2.0M
AI Moderator (×3 runs)	10.8–19.5m	✅	—
[aw] Failure Investigator	17.9m	❌	4.7M
Organization Health Report	13.8m	✅	1.3M
Schema Consistency Checker	13.1m	✅	8.1M
Daily CLI Tools Exploratory Tester	10.9m	❌	—

Three of the four slowest runs either failed or consumed very high tokens. Documentation Unbloat at 34.8 minutes is an outlier worth investigating — it ran Playwright and safeoutputs but still failed.

---

Resource Optimization Opportunities

18 workflows flagged HIGH severity resource_heavy_for_domain by the agentic profiler — the most frequent issue class this week. Notable examples:

Workflow	Turns	Agentic Fraction	Recommendation
Schema Consistency Checker	138	—	High — consider chunking or moving bulk reads to pre-steps
Contribution Check	30–51	0.50	Move ~50% data-gathering to deterministic pre-steps
Daily MCP Tool Concurrency Analysis	21	0.50	Move data-fetching to frontmatter steps
Go Fan	—	—	Review for unnecessary turns
Organization Health Report	—	—	Heavy but succeeded
jsweep	—	—	High for Code Fix domain

Workflows with agentic_fraction=0.50 have approximately half their turns doing pure data-gathering. The recommended fix in all cases is to move data-fetching to deterministic pre-agent steps (writing to /tmp/gh-aw/agent/) rather than having the agent issue repeated API calls itself. This reduces inference cost without changing task quality.

---

Recommendations

1. Fix Node.js PATH in chroot mode — Daily News and potentially other copilot-engine chroot workflows will continue failing until Node is resolvable post-pivot. Set GH_AW_NODE_BIN explicitly or add a toolcache Node to the chroot PATH. Severity: High (workflow completely non-functional).

2. Investigate add-shadow-ops docs build — The Build documentation step fails consistently on PR docs: Add CoprrectionOps #28676. Either fix the docs build on that branch or add a path filter to skip the step when docs sources haven't changed. Severity: Medium (PR-scoped, recovers when PR merges/closes).

3. Apply allowed-files fix to Schema Feature Coverage Checker — The agent succeeds but the PR is blocked by protected-files. Add allowed-files: [".github/workflows/schema-demo-*.md"] per Failure Investigator's recommendation in [aw] Schema Feature Coverage Checker failed #28671. Severity: Medium (blocker for workflow's core output).

4. Reduce Schema Consistency Checker token consumption — At 8.1M tokens and 138 turns per run with 14.9% cache efficiency, this workflow's weekly cost compounds quickly. Consider splitting into smaller batch jobs or pre-fetching schema data deterministically. Severity: Medium (cost).

5. Audit Documentation Unbloat — 34.8 minutes, 4.8M tokens, and still failing is an anomalous profile. The Playwright usage suggests it's loading external content. Add a timeout guard and review whether the Playwright step is necessary or can be replaced with a direct fetch. Severity: Low (single failed run, but pattern warrants review).

---

References:

• §24982458545 — [aw] Failure Investigator run (created [aw-failures] Failure Investigation Report — 6h window (2026-04-27) #28673, [aw-failures] Schema Feature Coverage Checker: protected-files config blocks all PRs from schema-demo-*.md outputs #28674)
• §24986870660 — Daily News node: command not found
• §24986681062 — Visual Regression Checker docs-build failure

Generated by Weekly Workflow Analysis · ● 238.6K · ◷

• expires on Apr 28, 2026, 10:07 AM UTC

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-04-28T10:07:05.002Z.

Closed by Workflow