disk-hygiene
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill makes extensive use of the
Bashtool to perform disk audits, measure directory sizes, and execute cleanup commands across various system and user directories (e.g.,rm -rf,brew cleanup,docker system prune). - [DYNAMIC_EXECUTION]: The instructions direct the agent to write multi-line bash scripts to temporary files in
/tmpusing theWritetool and then execute them usingbash. This technique is documented as a way to bypass shell environment hooks (likepueue) that might interfere with heredoc parsing. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by scanning the filesystem for filenames and directory structures (
find,du). An attacker could potentially influence agent behavior by naming files with instruction-like strings. This is partially mitigated by the use of theAskUserQuestiontool for user confirmation before any deletion. - [PRIVILEGE_ESCALATION]: The skill documentation includes troubleshooting advice that suggests the use of
sudoto delete root-owned system assets, such as macOS Aerial videos stored in/Library/Application Support/. - [PROMPT_INJECTION]: The skill contains a 'Self-Evolving' meta-instruction that directs the agent to immediately modify the
SKILL.mdfile using theWritetool if it encounters reproducible issues or drifts in parameters during execution.
Audit Metadata