The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to install the 'dust' utility on macOS by downloading a shell script from a personal GitHub repository ('https://raw.githubusercontent.com/bootandy/dust/refs/heads/master/install.sh') and piping it directly to the shell ('curl | sh'). This method of execution lacks integrity verification and is flagged as a high-risk pattern.
[COMMAND_EXECUTION]: The skill utilizes high-privilege system commands to perform audits and cleanup, including 'winget', 'brew', 'docker system prune', and 'wsl --shutdown'. It specifically checks for administrator privileges to enable high-speed NTFS scanning via MFT access, granting the agent significant control over the host operating system.
[EXTERNAL_DOWNLOADS]: The skill initiates the download and installation of external tools like 'WizTree' and 'disky' through platform-native package managers ('winget' and 'brew') to achieve its functional goals.
[PROMPT_INJECTION]: The skill design exposes an indirect prompt injection surface when processing file system data. 1. Ingestion points: File names and directory listings are read into the agent's context via 'Get-ChildItem', 'ls', and 'WizTree' CSV exports as defined in 'SKILL.md'. 2. Boundary markers: The skill emphasizes per-item manual confirmation and intent resolution to mitigate accidental actions. 3. Capability inventory: The agent has capabilities to delete files ('Remove-Item'), prune Docker data, and manage WSL instances. 4. Sanitization: While the instructions advise checking for symlinks and active processes, there is no automated sanitization of ingested file paths or names to prevent malicious strings from influencing the agent's logical flow.

disk-space-audit