Claude Skill

cleanup

Tidy the user's brain. Finds finished, forgotten or empty projects, duplicates, misplaced folders, loose files and old scratch, proposes what to archive, merge or move, and does only what the user approves. Use when the user types /cleanup, says "fais le ménage", "range mon brain

LLM Mart · 0 points · 1 views 0 listing impressions 0 install-command copies
Download Melvynx-ai-maxxing-skills-skills_brain_cleanup-86f0684.zip · 9 KB

Install

skills CLI npx skills add https://github.com/Melvynx/ai-maxxing-skills/tree/main/skills/brain/cleanup
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install melvynx-ai-maxxing-skills@llmmart
Git git clone https://github.com/Melvynx/ai-maxxing-skills.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole melvynx/ai-maxxing-skills collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Keep the brain small and true, so every session finds the right fact fast. The brain is the folder named in the ## Brain section of ~/.claude/CLAUDE.md (default ~/brain): read its AGENTS.md first (layout, hubs, hygiene). Talk to the user in their language, without jargon. Everything happens in the chat: no page, no report file. Never move, merge, edit or trash anything before the user says yes, and never delete permanently: archive what has value, trash only scratch. Never read or change archives/.

/dream merges duplicate lines every day inside the projects it touches. /cleanup is the monthly audit of everything: every project, every memory file, every skill.

  1. Scan. Run python3 <brain>/.agents/skills/cleanup/scripts/audit.py <brain> (read-only, up to a minute; --all lists every project and skill, --no-usage skips the reading of past sessions). When python3 is missing, collect the same facts yourself. It prints four parts. Each line is a lead, not a verdict: open the files it names before you propose anything.
  2. Audit. Give each lead one verdict: keep, update, merge, archive, trash, or ask (when only the user knows).
    • Projects. Archive: status done still in projects/; nothing changed for 60 days or more; paused for 30 days or more with no open task; an About.md still mostly "To fill". Merge: two projects about the same thing, a sub-project that repeats its hub. Move: a project at the projects/ root that belongs to a hub, or in the wrong hub. Fix: a project folder with no About.md, an About.md with no status: (a sub-folder that is only a part of a project needs neither: say so and leave it).
    • Memories (AGENTS.md, every About.md and Roadmap.md, notes/, the global rules). For each pair of files with duplicated sentences, read both passages: the same fact twice is a merge (keep one line in the narrowest place that covers every case, the newest wording, the oldest date, and a link where the other file needs it); a quote, a copy of an email or a template is kept. Two lines that disagree: the newer one wins when it clearly replaces the old; when nothing says which is right, look it up (the files, git history, the live tool), and ask only what no source settles. A heavy file: propose what leaves the current state (a finished launch becomes one line of result, a long history moves to files/ in the project, details in the brain AGENTS.md move to their About.md). A dead link: find the new path (git log --follow, search the name) and fix it, or remove the line when the thing is gone; a path that lives in another repo is not dead. A note untouched for 90 days: fold what is still true into a project, archive the rest. A rule in the brain AGENTS.md that repeats the global rules word for word: keep it in one place.
    • Skills (.agents/skills/). Not used in 60 days: ask whether it is still needed, then archive it to archives/skills/<name>/ (never trash a skill, and never touch one about security, recovery, money or a migration only because it is unused). Same name as a global skill, or two skills that describe the same job: keep the better one, fold in what the other adds. No description, a name that differs from its folder, a link to a missing file, not visible in .claude/skills: fix. Over 400 lines: move the detail to references/. Global skills are shown for information: only list them, a brain cleanup never changes them.
    • Tidy. Loose files or folders at the brain root that AGENTS.md does not list, empty folders, .tmp/ entries older than 7 days, report-like files (*-final*, *-v2*, handoff, summary) that repeat an About.md.
  3. Propose. One short numbered list in the chat, grouped by Projects, Memories, Skills and Tidy. One line each: what, the verdict, and the reason in a few words with the file and line when it helps ("editing-ia-pro/About.md L140: same Discord paragraph as ai-maxxing-pro/About.md L247, keep it in ai-maxxing-pro"). Leave out everything you verdict "keep". Say which numbers you recommend and ask in one message: "all", the numbers, or "none". When nothing needs doing, say the brain is clean and stop.
  4. Do only what they picked.
    • Archive a project: fold the final outcome into About.md (what was done, where the deliverables are), set status: done and updated: to today, trash scratch files, keep deliverables, and move the folder to archives/<folder-name>/ intact.
    • Merge: list every fact the source alone holds, write each one in the file that stays, then remove the lines (or archive the file left empty). A false merge is worse than a duplicate: merge only lines that say the same thing, and keep every detail one of them adds.
    • Move: move the folder into the right hub.
    • Fix: write the missing About.md from what the folder holds, with "To fill" where you would have to guess. Edit lines, never rewrite a whole file, and read a file again right before you edit it: another conversation may be saving at the same time.
    • Tidy: file each loose file into the project it belongs to, trash empty folders and old scratch, turn an old idea the user still wants into a project with the create-project skill, and trash the rest.
    • When the user says a flagged project or skill is still alive, set the project's updated: to today and leave it.
  5. Fix the links. Search the brain (and the global rules) for each path you moved or renamed and update it. Run the scan again: every lead you handled must be gone.
  6. Save. Commit only the files you changed with cleanup: <what was archived, merged, fixed>, and push when the brain has a GitHub copy.
  7. Report. Tell the user in 3 to 5 lines what changed (projects archived, duplicates merged, skills archived, space freed), how many projects are active now, what is left for them to decide, and that /cleanup once a month keeps it this way.
Files (ai-maxxing-skills)
  • scripts
    • audit.py 18.6 KB
      Loading…
  • SKILL.md 6.4 KB
    ---
    name: cleanup
    description: Audit and clean the whole brain. Checks the projects (finished, forgotten, misplaced), the memories (duplicates, facts that are no longer true, heavy files, dead links in AGENTS.md, About.md, Roadmap.md and notes), the skills (unused, duplicated, broken) and the loose files, shows one list of what to archive, merge, fix or trash, and does only what the user approves. Use when the user types /cleanup, says "fais le ménage", "range mon brain", "audite le cerveau", "audite les mémoires", "audite les skills", "archive les vieux projets" or the brain feels messy.
    metadata:
      internal: true
    ---
    
    Keep the brain small and true, so every session finds the right fact fast. The brain is the folder named in the `## Brain` section of `~/.claude/CLAUDE.md` (default `~/brain`): read its `AGENTS.md` first (layout, hubs, hygiene). Talk to the user in their language, without jargon. Everything happens in the chat: no page, no report file. Never move, merge, edit or trash anything before the user says yes, and never delete permanently: archive what has value, `trash` only scratch. Never read or change `archives/`.
    
    `/dream` merges duplicate lines every day inside the projects it touches. `/cleanup` is the monthly audit of everything: every project, every memory file, every skill.
    
    1. **Scan.** Run `python3 <brain>/.agents/skills/cleanup/scripts/audit.py <brain>` (read-only, up to a minute; `--all` lists every project and skill, `--no-usage` skips the reading of past sessions). When `python3` is missing, collect the same facts yourself. It prints four parts. Each line is a lead, not a verdict: open the files it names before you propose anything.
    2. **Audit.** Give each lead one verdict: keep, update, merge, archive, trash, or ask (when only the user knows).
       - **Projects.** Archive: status `done` still in `projects/`; nothing changed for 60 days or more; `paused` for 30 days or more with no open task; an `About.md` still mostly "To fill". Merge: two projects about the same thing, a sub-project that repeats its hub. Move: a project at the `projects/` root that belongs to a hub, or in the wrong hub. Fix: a project folder with no `About.md`, an `About.md` with no `status:` (a sub-folder that is only a part of a project needs neither: say so and leave it).
       - **Memories** (`AGENTS.md`, every `About.md` and `Roadmap.md`, `notes/`, the global rules). For each pair of files with duplicated sentences, read both passages: the same fact twice is a merge (keep one line in the narrowest place that covers every case, the newest wording, the oldest date, and a link where the other file needs it); a quote, a copy of an email or a template is kept. Two lines that disagree: the newer one wins when it clearly replaces the old; when nothing says which is right, look it up (the files, git history, the live tool), and ask only what no source settles. A heavy file: propose what leaves the current state (a finished launch becomes one line of result, a long history moves to `files/` in the project, details in the brain `AGENTS.md` move to their `About.md`). A dead link: find the new path (`git log --follow`, search the name) and fix it, or remove the line when the thing is gone; a path that lives in another repo is not dead. A note untouched for 90 days: fold what is still true into a project, archive the rest. A rule in the brain `AGENTS.md` that repeats the global rules word for word: keep it in one place.
       - **Skills** (`.agents/skills/`). Not used in 60 days: ask whether it is still needed, then archive it to `archives/skills/<name>/` (never trash a skill, and never touch one about security, recovery, money or a migration only because it is unused). Same name as a global skill, or two skills that describe the same job: keep the better one, fold in what the other adds. No description, a name that differs from its folder, a link to a missing file, not visible in `.claude/skills`: fix. Over 400 lines: move the detail to `references/`. Global skills are shown for information: only list them, a brain cleanup never changes them.
       - **Tidy.** Loose files or folders at the brain root that `AGENTS.md` does not list, empty folders, `.tmp/` entries older than 7 days, report-like files (`*-final*`, `*-v2*`, handoff, summary) that repeat an `About.md`.
    3. **Propose.** One short numbered list in the chat, grouped by Projects, Memories, Skills and Tidy. One line each: what, the verdict, and the reason in a few words with the file and line when it helps ("editing-ia-pro/About.md L140: same Discord paragraph as ai-maxxing-pro/About.md L247, keep it in ai-maxxing-pro"). Leave out everything you verdict "keep". Say which numbers you recommend and ask in one message: "all", the numbers, or "none". When nothing needs doing, say the brain is clean and stop.
    4. **Do** only what they picked.
       - Archive a project: fold the final outcome into `About.md` (what was done, where the deliverables are), set `status: done` and `updated:` to today, trash scratch files, keep deliverables, and move the folder to `archives/<folder-name>/` intact.
       - Merge: list every fact the source alone holds, write each one in the file that stays, then remove the lines (or archive the file left empty). A false merge is worse than a duplicate: merge only lines that say the same thing, and keep every detail one of them adds.
       - Move: move the folder into the right hub.
       - Fix: write the missing `About.md` from what the folder holds, with "To fill" where you would have to guess. Edit lines, never rewrite a whole file, and read a file again right before you edit it: another conversation may be saving at the same time.
       - Tidy: file each loose file into the project it belongs to, trash empty folders and old scratch, turn an old idea the user still wants into a project with the `create-project` skill, and trash the rest.
       - When the user says a flagged project or skill is still alive, set the project's `updated:` to today and leave it.
    5. **Fix the links.** Search the brain (and the global rules) for each path you moved or renamed and update it. Run the scan again: every lead you handled must be gone.
    6. **Save.** Commit only the files you changed with `cleanup: <what was archived, merged, fixed>`, and push when the brain has a GitHub copy.
    7. **Report.** Tell the user in 3 to 5 lines what changed (projects archived, duplicates merged, skills archived, space freed), how many projects are active now, what is left for them to decide, and that `/cleanup` once a month keeps it this way.
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.