The Agent Skills Directory

[SAFE]: The skill utilizes the keyring library to store OAuth tokens in the system's secure credential store rather than in plain text configuration files.
[SAFE]: Implements a structured permission model that classifies commands into read and write categories, explicitly instructing the agent to seek user confirmation for any actions that create, modify, or delete data.
[SAFE]: Uses official Google API client libraries and standard Python argument parsing to ensure secure communication and safe handling of command-line inputs.
[PROMPT_INJECTION]: The skill allows the agent to ingest external data from Google Drive, creating a surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent's context through file names, folder names, and metadata retrieved via the files and folders commands in scripts/google-drive.py.
Boundary markers: The skill documentation does not provide specific delimiters or 'ignore' instructions for the model when processing retrieved file data.
Capability inventory: The skill provides the agent with multiple write capabilities, including files upload, files move, files delete, and the share command in scripts/google-drive.py.
Sanitization: While shell inputs are sanitized via the argparse library, the skill does not perform semantic filtering of the file metadata returned by the Google Drive API.

google-drive