The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends and automates the installation of 'mole', a third-party command-line utility for macOS disk cleanup, from a Homebrew tap (tw93/tap/mole).
Evidence: Installation and update instructions are found in SKILL.md and references/mole_integration.md.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted local filesystem metadata, such as file and directory names, which could potentially contain malicious instructions intended to influence the AI agent's behavior.
Ingestion points: Local filesystem metadata is read using os.scandir, find, and du across several scripts (e.g., scripts/find_app_remnants.py, scripts/analyze_large_files.py).
Boundary markers: SKILL.md contains explicit 'ABSOLUTE PROHIBITIONS' and strict instructions for a 'safety-first philosophy', mandating that the agent never auto-execute deletions and always require user confirmation.
Capability inventory: The skill possesses file deletion capabilities through shutil.rmtree and os.unlink in scripts/safe_delete.py.
Sanitization: Input paths are handled as literal strings in Python file operations and list-based subprocess calls, which prevents shell command injection. Metadata such as application names are normalized to alphanumeric characters before processing in scripts/find_app_remnants.py.
[COMMAND_EXECUTION]: The skill executes various macOS system utilities (du, df, find, brew, docker) through Python's subprocess module.
Evidence: Found across all scripts in the scripts/ directory. These calls are implemented using list-based arguments and avoid shell=True, which is a secure implementation practice.
Note: There is a minor inconsistency where scripts/analyze_dev_env.py suggests a Docker cleanup command (docker system prune -a --volumes) that is explicitly prohibited by the safety rules in SKILL.md.

macos-cleaner