Claude Skill Code Cleanup
Two skills for Codex and Claude Code: a read-only repository audit and an experiment loop with saved baselines.
- 88 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Reads the repository and runs check scripts on your machine
- The experiment mode may change code after your permission
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/claude-skill-code-cleanupDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Reads the repository and runs check scripts on your machine
- The experiment mode may change code after your permission
This is third-party code. Review the repository files before installing.
What it does
The repository holds two skills. `code-cleanup-helper` audits code, architecture, repository, release and the skills themselves without changing anything: it looks for duplicates, naming drift, dependencies and long functions and checks security, token use and updates. Each check is marked `FAIL`, `REVIEW`, `NOT_CHECKED` or `PASS`, and anything unmeasured cannot be passed off as passed. `run-benchmark-driven-rd` sets a falsifiable goal, saves a baseline, runs experiments and decides what to accept, with control over external changes. Reports use JSON, and the `context_intake.py` script chooses which reference files to read at startup and after context compaction without loading everything at once.
Who it is for. Engineers who trust an agent with repository checks and want evidence in place of general claims of success.
Good fit when
- Before a release, a repository check is needed in which every category is explicitly marked passed or not checked.
- A change should be accepted only after measuring against a saved baseline.
- A skill or skill set needs a check for drift, permissions and wasted tokens.
- An agent overloads its context and must choose which reference files to load at start.
Not a fit when
- A quick code edit is needed with no reports: the audit adds procedure and output.
- The project is small and throwaway: the skills carry a heavy check contract.
- An automatic updater is required: the skills do not run in the background and scan nothing on their own.
Example request
Audit the repository without changes and mark every category as PASS, REVIEW or NOT_CHECKED.Limitations
Documentation is mostly in Chinese (traditional script), and the skills themselves contain Chinese text. Installation is manual: clone the repository, run `python -m pip install -r requirements.txt` and copy both folders into `~/.codex/skills/` or `~/.claude/skills/` side by side, otherwise R&D cannot find the cleanup provider. The version at the time of review is 0.24.0. The author marks benefit claims as provisional, and passing tests does not prove overall agent quality.
How to disable. Delete both skill folders from `~/.claude/skills/` or `~/.codex/skills/` and open a new session. Local report copies are erased by hand.
FAQ
Does the audit change files?
No, `code-cleanup-helper` is read-only. It issues a diagnosis first and waits for permission, and edits happen separately on your word.
What do the check statuses mean?
`FAIL` means a proven blocker, `REVIEW` needs a human decision, `NOT_CHECKED` shows a measurement gap and `PASS` covers only the checks that were run.
Are both skills required together?
They must sit side by side. `run-benchmark-driven-rd` uses `code-cleanup-helper` as a cleanup provider and cannot find it otherwise.
Does it run in the background?
No. The skills do not scan projects on a schedule or rewrite code themselves, and take part only in a development session.
How do I check context loading?
By running `python code-cleanup-helper/scripts/context_intake.py run-benchmark-driven-rd --event startup --emit-packet` at the repository root. Output is JSON.
Discussion
No questions or reviews yet.
Related
SkillsThe evidence-code-review skill for Codex and Claude Code: evidence-based change review with MUST, SHOULD, BETTER and NITS levels
Eleven skills to verify AI-written code: change blast radius, defect scan, review report, test cases and browser flow replay
Four Claude Code skills: 1C testing through Codex and 1c-testpilot, code review, export and load through ibcmd
A CLI and MCP for testing agent skills: repeatable YAML evals, report comparison, regression detection and skilled-improve
Runs Claude Code, Codex, Gemini and OpenCode on YAML tasks in a sandbox, scores the result and gates CI on it
/pr-review sends a pull request through five independent reviewers, cross-checks findings and builds one report.
README in short
Describes the two skills, version changes from 0.22 to 0.24, manual installation for Codex and Claude Code, upgrades and CI checks on three operating systems.