The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, creating an attack surface for malicious instructions to influence the agent's behavior.\n
Ingestion points: In scripts/google_api.py, the skill fetches content from Gmail messages (gmail get), Google Docs (docs get), and Drive files.\n
Boundary markers: There are no explicit boundaries or instructions to the agent to treat fetched content as data rather than instructions.\n
Capability inventory: The skill possesses significant capabilities that could be misused, such as sending emails (gmail send), modifying labels, creating/deleting calendar events, and updating spreadsheets.\n
Sanitization: No sanitization is implemented to filter or escape instructions embedded within the retrieved messages or documents.\n- [EXTERNAL_DOWNLOADS]: The skill installs necessary Google API libraries through the setup script.\n
Evidence: scripts/setup.py uses subprocess.check_call to install google-api-python-client, google-auth-oauthlib, and google-auth-httplib2 via pip.\n- [COMMAND_EXECUTION]: The skill uses an external CLI tool to handle Workspace operations when available.\n
Evidence: scripts/google_api.py and scripts/gws_bridge.py execute the gws binary using subprocess.run to interact with Google Workspace APIs.

google-workspace