agents-md-improverlisted

# AGENTS.md Improver Audit and improve agent context files so they are **slim, verified, and worth their context cost**. ## File layout this skill enforces - **`AGENTS.md`** is the canonical context file. All project guidance lives here. - **`CLAUDE.md`** is a thin pointer that imports it. This isn't just tidiness: Claude Code reads `CLAUDE.md`, not `AGENTS.md` — the pointer is what makes AGENTS.md load at all (and it's the layout the official memory docs recommend for repos shared with other coding agents): ```markdown # CLAUDE.md @AGENTS.md ## Claude Code (only Claude-specific guidance, if any — otherwise omit this section) ``` Enforce this layout in every audit, whatever the starting state: | Starting state | Action | |---|---| | AGENTS.md only | Audit it, **and create the pointer CLAUDE.md** | | Full CLAUDE.md only | Migrate content → AGENTS.md, replace CLAUDE.md with the pointer | | Both, duplicated | AGENTS.md becomes canonical, CLAUDE.md becomes the pointer | | Neither | Build a slim AGENTS.md from *verified traps actually found*, plus the pointer. If the repo has no traps, a near-empty file (one-line description + package manager if non-npm + non-obvious build command) is the correct output — never pad to look complete. | Never maintain the same content in both files. Creating the pointer CLAUDE.md is part of the standard output, not an optional extra — include it in the Phase 4 report. ## Core principle Every line is loaded int