Project Artifact Cleanup — AI skill for Claude Code

Project Artifact Cleanup

Stop AI-generated scratch files from piling up. A Codex and Claude Code skill with deletion previews, configurable retention, pinning, and recovery before expiry.

See it work

Run a self-contained demo from this repository with Python 3.9+:

python3 scripts/demo.py

It uses the real CLI with generated sample files in a temporary directory:

1. Created one 13-byte scratch file in a managed task; app.py stays outside.
2. Zero-day preview: would_delete_bytes=13, payload_deleted=false.
3. Retained for 7 days; sweep skips the task because it has not expired.
4. Recovered identical bytes; original pinned and skipped by the next sweep.
5. app.py unchanged. PASS. Removing only this demo's temporary fixtures.

The demo checks each result and exits with an error if it differs. It does not scan an existing project. Its generated fixtures are removed when it exits. This demonstrates the helper's behavior; deciding which real files are disposable still requires your judgment. [Run the commands yourself](references/setup.md).

Install

From your project directory, choose the command for your agent. Existing destinations are not overwritten by `git clone`.

**Codex:**

git clone https://github.com/ShiYuPro/project-artifact-cleanup.git .agents/skills/project-artifact-cleanup

**Claude Code:**

git clone https://github.com/ShiYuPro/project-artifact-cleanup.git .claude/skills/project-artifact-cleanup

Invoke `$project-artifact-cleanup` in a new task. Discovery depends on your host's support for `SKILL.md`; installation does not change project policy or grant external permissions.

First use

Use $project-artifact-cleanup to preview agent-created artifacts in this project. Do not delete anything yet.

Run helper commands from the installed skill directory or this repository root. See [setup and examples](references/setup.md) and the [full skill](SKILL.md).

Requirements

Python 3.9+ on macOS, Linux or