TsukiNi22/ai-utils / skills/system/computer-clean
computer-clean
UnverifiedFree as much space as possible on the user's computer and in the data of Claude Code / claude-cluster - a deep scan of every ~/.cache folder, the caches of the language tools and of the Chromium / Electron / flatpak apps, build outputs and envs of old projects (never a git-tracked file), the trash, downloads (installers, archives already extracted), duplicated or big old files, core dumps, Docker / Podman (images, orphan volumes), flatpak (unused runtimes, SDKs, data of removed apps), dnf (cache, updates not installed), journals, old kernels, old Claude Code conversations, finished jobs, clean worktrees, claude-cluster leftovers - each with its size (real gain on btrfs), what it is for and what deleting it does, chosen in a QCM, then deleted for good. Use whenever the user wants to clean / free space / tidy their computer, their disk, their caches, Docker, or the leftovers of Claude Code / claude-cluster (old sessions, jobs, worktrees).
Computer clean
SKILL_DIR = the directory of this file. The script finds, the user chooses, the script deletes: nothing is deleted
that the user did not tick. Deletion is for good (no trash: the user's choice), so the list must be right and each
item explained: what it is for, what deleting it does.
1. Find
python3 SKILL_DIR/scripts/computer_clean.py scan [--days 30] [--min-mb 5] [--root ~/projects]... [--top 12] [--json]
Run df -h ~ / first (the "before" of the report). A deep clean: --days 7. Each candidate has an id, its size, its
age, safe (made again when needed) or review (to check), for (what it is for), after (what deleting it
does), its command when it is not a plain rm -rf, and on btrfs frees at least (the part no other file shares:
uv venvs, cp --reflink copies and snapshots keep the shared part alive - ex: a 6 GB uv cache that frees 45 MB).
Five groups:
- Caches: every folder of
~/.cache(known onessafewith their own text, unknown onesreview: some tools keep real data there), the caches of the language tools (~/.npm, pnpm store viapnpm store prune, cargo,go clean -modcache, gradle, maven, nuget, conan, yarn, bun,~/.debugof perf, Android, PlatformIO, pipenv envs, ollama models), the web caches of the Chromium / Electron apps in~/.config(Discord, VS Code, Chrome...) and the caches of the flatpak apps (~/.var/app/*/cache). - Build outputs and envs of old projects under
--root(default the home folder; hidden folders and installed programs skipped):build,build-*,cmake-build-*,node_modules,target,.venv/venv(review),dist(review),.next,__pycache__,.pytest_cache,.gradle, zig / dune / stack / cabal / .NET outputs... each next to its project file, untouched for--days, never tracked by git. - Files: the trash, the downloads folder of the desktop (
XDG_DOWNLOAD_DIR:~/Téléchargementsin French) - archives already extracted next to their folder, installers and disk images (.iso,.rpm,.deb,.AppImage...), big downloads (> 200 MB) - big files present twice in the desktop folders (same content, every copy but one), big files untouched for months (> 500 MB, 90 days), core dumps (core,vgcore.*), big logs (> 50 MB). - System and containers: Docker / Podman (
system prune -af: unused images, stopped containers, build cache) and their orphan volumes (one by one: the data of an old compose project), the user journal and the system journal (2 weeks kept), dnf cache (/var/cache/libdnf5,/var/cache/dnf), the GNOME Software cache, updates downloaded but not installed (/var/lib/dnf/offline, often GBs), Fedora upgrade leftovers, system crash dumps, old kernels (never the running one nor one fallback), flatpak runtimes no app uses, flatpak SDKs no app runs on (--unusedkeeps them), data of flatpak apps no longer installed. Root ones say so. - Claude Code / claude-cluster: conversations of
~/.claude/projectsolder than--days, finished jobs of~/.claude/jobs(never the running one), old shell snapshots / file history / todos, worktrees of.claude/worktreeswith nothing uncommitted (git worktree remove, the branch kept), claude-cluster: transcripts of sessions it no longer has,logs/archive, the showcase session, the debug trace, old pasted pictures.
Last, the biggest folders of the home (information only, never candidates): what is big but not a leftover (AI
models, venvs of tools, wine prefixes, projects). --json gives the candidates to read programmatically.
2. Show and ask (AskUserQuestion, French, interactive-questions rules)
- First the summary in text:
df(used / free), the total per group, the biggest items with their real gain on btrfs, what isreviewand why (ex: "un volume Docker orphelin = la base de données d'un ancien projet compose"), and the biggest folders worth knowing about. - Then the QCM (multiSelect), 1 to 4 questions of up to 4 options, by theme: Caches, Builds, Fichiers,
Système / conteneurs, Claude; each option = one candidate or a set of the same kind (label: what + size, or the
real gain when it is much smaller; description: what it is for and what deleting it does, from
for/after). Thesafeones recommended, thereviewones explained. More candidates than fit: several rounds, or a set per kind ("18 dossiers target de Rust de la piscine, 180 Mo"). - Updates waiting to be installed: recommend installing them (restart and update in GNOME Software: they go by themselves) when there is room, deleting them when the disk is full.
- No answer for a group: nothing of it is deleted. A question about an item ("c'est quoi uv ?"): answer it (what it is, what uses it, the real gain) and ask again before deleting it.
3. Delete the chosen ones
python3 SKILL_DIR/scripts/computer_clean.py clean <id> <id>... [--days 30 ...] # the same options as the scan
The scan is made again with the same options: an id that changed or disappeared is skipped (said). Before a review
one, look at it: a Docker volume (docker volume inspect: compose project, date; which volume the running container
mounts), a node_modules / venv mounted in a running container (docker inspect -f '{{.Mounts}}'), a download or a
big file (its name, what folder it belongs to). Root ones (dnf, journals, crash dumps, kernels, PackageKit):
SUDO_ASKPASS=~/.local/bin/sudo-askpass exported, run with sudo -A (a password window on this computer); without it,
the command to run is printed. Flatpak SDKs are removed with --no-related (their related extensions, like
codecs-extra, may be needed by a Platform). --dry-run first when the selection is big.
4. Report
A short table: what was deleted, the space freed (df -h ~ before / after; on btrfs the space can come back a few
seconds later), what failed and why, what was left (unticked, review). Never delete anything outside the list of the
scan; a request beyond it (a project folder, a whole directory, one of the biggest folders): say it is not part of
this skill and ask explicitly.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 318k | Updated | Claude CodeCodex | Unverified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 117k | Updated | Claude CodeCodex | Unverified | Install |