The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the gog CLI tool to automate tasks across Google Workspace. It implements several security best practices, including the use of --dry-run for testing, --no-input for headless environments, and explicit command filtering via --enable-commands and --disable-commands to enforce the principle of least privilege.
[PROMPT_INJECTION]: The skill acknowledges the risk of indirect prompt injection when reading untrusted data from emails, documents, or spreadsheets.
Ingestion points: SKILL.md (via data-fetching commands such as gog gmail get, gog docs cat, and gog drive ls).
Boundary markers: Present; the skill explicitly instructs the agent to use the --wrap-untrusted flag when reading Google content to provide clear delimiters for the LLM.
Capability inventory: SKILL.md (includes mutation commands like gog docs write, gog sheets update, and gog drive upload across the implementation files).
Sanitization: Present; the skill recommends the --sanitize-content flag to strip potentially malicious payloads from Gmail message bodies before processing.
[SAFE]: The skill demonstrates high security maturity by providing explicit warnings against printing access tokens or secrets, advising on secure keyring configurations, and encouraging the use of scoped readonly access whenever possible.

gog