The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the gog (gogcli) binary to interact with the Gmail API. It executes commands for searching threads, retrieving messages, managing labels, and sending emails.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for the user to install an external dependency via Homebrew (brew install gogcli). This is a guided manual step rather than an automated silent download.
[DATA_EXPOSURE]: The skill processes sensitive data including email bodies, subjects, and attachments. This is consistent with its primary purpose. It advises users to handle OAuth client_secret JSON files manually and stores credentials in the tool's native keyring rather than the skill directory.
[PROMPT_INJECTION]: No evidence of malicious prompt injection, behavioral overrides, or safety filter bypasses was found in the instructions.
[INDIRECT_PROMPT_INJECTION]: The skill has an ingestion surface for untrusted data by reading email bodies (e.g., gog gmail messages search --include-body).
Ingestion points: Reads email content from Gmail threads and messages in SKILL.md and references/gog-gmail-commands.md.
Boundary markers: None explicitly defined in the CLI output processing, but results are requested in --json format for structured parsing.
Capability inventory: Can send emails, create drafts, and modify labels/filters.
Sanitization: The skill mitigates risks by explicitly requiring the agent to confirm recipients, subjects, and bodies with the user before any 'send' operation is executed.

gmail