review-file-size

npx skills add majiayu000/claude-skill-registry --skill review-file-size --agent claude-code

Same command for any agent — swap --agent for codex, cursor, copilot.

Enter planning mode. Identify source files approaching the context load limit and evaluate whether they can be meaningfully split. Use parallelism where possible. ## Why This Matters Files over ~25k tokens cause problems: the Read tool must batch them, they bloat context for tasks that only need a small piece, and they're harder to reason about in a single pass. But splitting a file poorly — into two halves that share all the same globals and must always be loaded together — is worse than one large file. A split only helps if the pieces are **independently useful**. ## Step 1 — Measure file sizes Scan all `.ahk` files in `src/` and `tests/` (excluding `src/lib/`). For each file, estimate the token count — don't rely on line counts alone. AHK files with long DllCall signatures, inline strings, or dense logic have significantly more tokens per line than sparse files with short functions. **Approximate token estimation**: Read each file and estimate tokens. Characters ÷ 4 is a rough baseline, but AHK identifiers and camelCase inflate this. Aim for a reasonable estimate, not exact counts. **Threshold**: Only flag files that are **approaching trouble** — roughly 20k+ tokens. Files well