The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from Google Workspace APIs (Gmail, Docs, Drive) that may contain malicious instructions designed to influence the agent's behavior. * Ingestion points: Content is ingested through commands such as gmail get, docs cat, and sheets get as described in SKILL.md. * Boundary markers: The documentation suggests using the --wrap-untrusted flag to identify and isolate external data. * Capability inventory: The CLI tool has significant capabilities including sending emails, modifying documents, and managing authentication state. * Sanitization: The skill recommends the --sanitize-content flag to clean Gmail message bodies before processing.
[DATA_EXFILTRATION]: The skill provides instructions for managing sensitive authentication data and Google service access. * Evidence: SKILL.md describes procedures for capturing OAuth consent URLs and sourcing GOG_KEYRING_PASSWORD from the environment. * Mitigation: The instructions explicitly forbid printing access tokens or secrets and advise using restricted file permissions (mode 0600) for sensitive temporary files.

gog