The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data such as filenames and tool outputs during disk scanning. Mitigations include explicit instructions to treat this data as non-executable, the use of quoted literal arguments, and the recommendation of NUL-safe enumeration to prevent command injection. Evidence found in SKILL.md and references/scan.md.- [COMMAND_EXECUTION]: Disk analysis and cleanup are performed via shell commands. Safety is maintained by forbidding sudo, requiring explicit user authorization for deletions, and verifying tool versions before use. The skill also warns against the use of auto-installing shims or tools with unknown side effects. Evidence found in SKILL.md and references/cleanup.md.- [DATA_EXPOSURE]: The skill contains strict prohibitions against reading sensitive data like .env files, private keys, messages, or browsing history. It also ensures that all scan results and inventories are kept on the local system and not sent to external services. Evidence found in SKILL.md.

mac-space-cleanup