The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bundled Python script (scripts/skill_cleaner.py) to perform filesystem scanning, read file contents, and manage the quarantine area. It uses standard library functions like shutil.move and os.unlink to isolate or restore skills based on user instructions.
[PROMPT_INJECTION]: The skill ingests and displays content from external, potentially untrusted agent skills to generate security reports, which introduces a surface for indirect prompt injection.
Ingestion points: The script reads the contents of SKILL.md and executable files within directories such as ~/.claude/skills and ~/.codex/skills.
Boundary markers: The reporting output uses a structured Markdown template, but it lacks explicit instructions telling the agent to disregard commands found within audited content snippets.
Capability inventory: The skill can move directories and remove symbolic links within the user's agent skill directories.
Sanitization: Content is read and displayed as text excerpts without specific filtering or escaping for prompt injection sequences.
[SAFE]: The skill's behavior is consistent with its stated purpose as a security auditing utility. It performs all operations locally, shows no network activity, and requires explicit confirmation flags (e.g., --yes) before performing any destructive or isolation-based file operations.

dbs-skill-cleaner