The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes powerful shell commands to perform disk cleanup, including rm -rf for deleting directories and specific cleanup flags for package managers like brew cleanup, uv cache clean, npm cache clean, and pip3 cache purge. These operations are performed on standard cache locations and are consistent with the skill's primary purpose.
[INDIRECT_PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting the output of du (disk usage) and ls commands. An adversary could potentially create directories with malicious names intended to influence the agent's subsequent logic during the analysis phase.
Ingestion points: Output of du commands used to measure top-level and home directory usage in SKILL.md, and ls output for Python versions in references/paths.md.
Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded content in directory names when processing command output.
Capability inventory: The agent has the ability to delete files and directories via rm -rf, prune Homebrew caches, and remove Ollama models through the CLI.
Sanitization: The skill mitigates risk by instructing the agent to classify candidates into 'quick wins' (caches) and 'confirm-first' groups, mandating individual user approval for the latter.

macos-disk-cleanup