agents-md

To audit AGENTS.md, run a quick or full audit workflow that scores signal-to-noise, checks for stale commands, and proposes diffs to enforce execution-first guidance. This prunes bloated sections and normalizes instructions.

Execution-first guidance prioritizes actionable commands, gotchas, and concise rules over verbose explanations in agent instruction files. Auditing AGENTS.md enforces this by pruning bloated content and ensuring the file remains the source of truth.

Refactor AGENTS.md by running an audit workflow that identifies stale commands and bloated sections, then applies proposed diffs to normalize guidance. This ensures instructions stay concise, actionable, and execution-first.

Yes, the audit workflow includes guidance normalization to ensure AGENTS.md remains the source of truth and correctly symlinks to CLAUDE.md. This maintains consistency across agent instruction files.

A quick audit identifies high-priority fixes in AGENTS.md using provided references, while a full audit uses comprehensive reference materials to score and propose diffs across all sections. Both enforce execution-first guidance.

You should avoid a full audit when your AGENTS.md only needs high-priority fixes, as the quick audit mode is faster. Use the full audit only when you need comprehensive scoring and diff proposals across all reference materials.