github/recompile-workflows

Recompile Agentic Workflows

Use this skill when you need to regenerate all agentic workflow lock files and apply post-processing.

IMPORTANT: Post-processing is required after EVERY lock file change

Any time .lock.yml files are regenerated — whether via gh aw compile, gh aw upgrade, or any other gh-aw command — you MUST run the post-processing script afterward. This is not optional.

Steps

1. Compile or upgrade workflows

Use whichever command is appropriate:

# Full upgrade (updates agents, actions, codemods, then compiles)
gh aw upgrade

# Just recompile (when only .md workflow files changed)
gh aw compile

If any workflow fails to compile (e.g., strict mode violations like contents: write), fix the .md source file and re-run.

2. Run post-processing script (ALWAYS)

This step MUST run every time lock files are regenerated, regardless of how they were generated.

The post-processing script replaces the "Install awf binary" step in smoke and build-test workflows with local build+install steps, so CI tests the repo's own code instead of a released binary.

npx ts-node scripts/ci/postprocess-smoke-workflows.ts

This updates these lock files:

• smoke-copilot.lock.yml
• smoke-claude.lock.yml
• smoke-chroot.lock.yml
• build-test.lock.yml

Common Issues

Strict mode violations

Newer gh-aw versions enforce strict mode which disallows write permissions like contents: write, issues: write, etc. Workflows should use safe-outputs for write operations and only request read permissions.

Discussion category warnings

Warnings about "General" vs "general" discussion category casing are non-blocking.

Verification

After both steps, run git diff --stat to review all changed files. Expect changes in:

• .github/agents/ - Updated agent files
• .github/aw/actions-lock.json - Updated action pins
• .github/workflows/*.lock.yml - Regenerated lock files
• .github/workflows/*.md - If codemods applied fixes