Clean My Disk

clean-my-disk

[](https://github.com/richer-richard/clean-my-disk/actions/workflows/validate.yml)

**A safe, free disk-cleanup skill for [Claude Code](https://claude.com/claude-code).** Reclaim disk space on **macOS, Linux, and Windows** — without paying for CleanMyMac, DaisyDisk, or BleachBit. Install it once, then just tell Claude *"my disk is full"*.

It is built on one principle paid cleaners get right and most `rm -rf` scripts get wrong: **measure first, delete only what is known to be disposable, and show you the bill.** It scans **read-only**, reports what it *could* reclaim grouped by risk tier, then deletes **only** an allowlist of regenerable caches, temp files, and trash. Your documents, keys, app data, sync folders, VM images, and source trees are never touched.

---

Why

Agents with delete permission are powerful and dangerous. "Find big files and delete them" is exactly how you lose irreplaceable data. This skill encodes the restraint a good cleaner has:

• Allowlist, not heuristics. It removes only enumerated cache/temp/trash locations. Big ≠ deletable. Large files are reported, never auto-removed.
• Native cleaners first. brew cleanup, npm cache clean, docker builder prune, Clear-RecycleBin, etc. understand their own data and fail safe. Raw rm is the last resort, and only ever on cache contents.
• A hard protected list. Credentials, Application Support, iCloud/Dropbox/OneDrive sync roots, VM disk images, Docker volumes, and .git/source trees are off-limits, always — even under an "autonomous" or bypassPermissions run.
• Measure → plan → clean → report. You always get a disk-cleanup-report.md with per-category bytes freed and a list of everything skipped and why.
• Never deletes on a name match. A substring match is a hypothesis, not an identification. *ace* once matched four unrelated produ