The Agent Skills Directory

[COMMAND_EXECUTION]: The archive.py script performs significant file system modifications, including shutil.move and Path.rename operations across the user's ~/.claude and ~/.codex directories. While it includes protections against modifying its own archive or plugin cache, it has the capability to move arbitrary directories within the scope of skill locations.
[INDIRECT_PROMPT_INJECTION]: The skill relies on parsing an HTML report (skill-stats.html) generated by another utility. This report contains a JSON-like array (ROWS) containing skill names and file paths. If an attacker can influence the contents of this report (e.g., by creating a skill with a malicious name or path), they could potentially cause the cleanup script to move unintended files or directories when --apply is used. However, this is mitigated by the fact that the report is intended to be generated locally by a related trusted tool.
[SAFE]: The skill implements several safety features, including a mandatory --apply flag for any changes (defaulting to dry-run), preventing operations on sensitive plugin cache directories, and ensuring the archiving process is fully reversible by renaming SKILL.md to SKILL.md.archived rather than deleting it.

skill-cleanup