The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to read shell configuration files (such as ~/.zshrc and ~/.bashrc) to check for directory references in the system PATH. While these files often contain sensitive secrets like API keys, the skill includes explicit instructions to redact any such secrets before they are recorded in logs or shared in chat.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from the local filesystem, including project folders, downloads, and notes. This constitutes an ingestion point for untrusted data that could influence agent behavior.
Ingestion points: Local project directories (~/Documents/GitHub, ~/projects), user notes (<NOTES_ROOT>), and the downloads folder (SKILL.md).
Boundary markers: No specific delimiters or boundary markers are used when triaging file content.
Capability inventory: System inventory tools (du, ls, grep), package managers (brew, npm), and file management (mv, mkdir).
Sanitization: The instructions require the agent to redact matching secret lines found during shell config inspection.
[COMMAND_EXECUTION]: The skill executes multiple system commands to audit disk usage, check for active services, and inventory package managers. It follows a "review-only" default mode and uses the macOS Trash for all deletions to ensure recoverability.
[DYNAMIC_EXECUTION]: A Python script is used locally to estimate the token count of the agent's main instruction file. This is a hardcoded utility function for budget management.

hygiene