[github-actions]

Audit summary

Release and change audit: Latest release is v0.93.2 (published 2026-10-23T05:57:09Z). The repository HEAD (764fdc6) is the exact commit the release was cut from, so there are zero commits touching docs/ or .github/aw/ since the release. No content-accuracy changes were needed on that basis.

Size audit (wc -l .github/aw/*.md): one violation found — debug-workflow-evidence.md was 453 lines, over the 400-line limit for .github/aw/*.md files. Because this violation exists, the edge case that allows exiting immediately (no doc changes and all files under limit) did not apply, so I also ran the duplication and accuracy audits below.

Fix applied: trimmed debug-workflow-evidence.md from 453 → 390 lines by consolidating duplication, without moving any heading to another file:

• Removed the ## Quick Start section — it fully duplicated the Basic Usage examples shown a few lines later under Downloading Workflow Logs / Auditing Specific Runs; its one unique instruction (keep URLs intact, qualify bare IDs with --repo) was folded into the Downloading Workflow Logs intro instead of being dropped.
• Condensed the Search and Cluster Session Evidence caveat paragraph, which restated detail already owned canonically by analyzing-runs.md#retrieve-and-cluster-session-evidence.
• Collapsed four single-command "Advanced Debugging Techniques" subsections (Checking Workflow Status, Downloading Specific Artifacts, Inspecting Job Logs, Analyzing Firewall Logs) into one Other Useful Commands block.
• Collapsed the Reference Commands section's four subsections (each just a sentence pointing at a cli-commands.md anchor already demonstrated in-file) into one short paragraph.
• Tightened the Check Whether an Error Recurs paragraph and shortened a multi-line jq example to one line.

I deliberately did not move any section to a new file. .github/skills/agentic-workflows/tests/test_guidance.py pins several exact headings/substrings to this specific file via EVIDENCE_GUIDE.read_text().split(...) (e.g. ### Missing Tool Errors, ### Diagnostic and Development Compilation, the Compilation Process dry-run phrasing, and the preamble before ## Table of Contents). I cross-checked every pinned requirement by hand against the edited file and left all of that content byte-for-byte unchanged. I could not run the Python test suite myself (my sandbox restricts Bash to the audit's declared commands — git log, git diff, wc -l — so a cross-file split that would've required coordinated edits across that test file wasn't something I could safely verify).

Duplication audit: spot-checked the "single-job constraint" pattern across workflow-constraints.md, workflow-patterns.md, and create-agentic-workflow.md — already well-factored (one canonical statement in workflow-constraints.md, short cross-references elsewhere). No action needed there.

Accuracy audit: confirmed pkg/workflow/compiler_types.go and pkg/parser/schemas/main_workflow_schema.json still exist as named. Note pkg/workflow/safe_outputs_config.go referenced in AGENTS.md's audit checklist has since been split into several files (safe_outputs_config_base.go, safe_outputs_config_types.go, etc.) — the underlying safe-output behavior documented in .github/aw/safe-outputs*.md still matched what I spot-checked, but the janitor's own source-of-truth file list is now slightly stale (out of scope for this PR, which only touches .github/aw/*.md).

Files changed

File	Before	After
.github/aw/debug-workflow-evidence.md	453	390

All other .github/aw/*.md files were confirmed under their target limits.

🤖 Generated with [Claude Code]((claude.com/redacted)

Generated by 🧹 Instructions Janitor · claude · sonnet50 · ~179 AIC · ⌖ ~19 AIC · ⊞ 6.7K · ◷

• expires on Oct 25, 2026, 12:51 AM UTC-08:00