The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions involve downloading the gogcli binary from GitHub (openclaw/gogcli). The source is a well-known service and the version is pinned for verification.
[COMMAND_EXECUTION]: The agent is authorized to execute gog commands via the Bash tool (gog:*) to interact with Google Workspace services.
[PROMPT_INJECTION]: The skill processes untrusted content from Google Workspace, creating a surface for indirect prompt injection.
Ingestion points: Data is ingested via gog gmail get, gog drive download, gog docs cat, and gog sheets get (specified in SKILL.md and referenced documentation).
Boundary markers: The instructions do not provide explicit delimiters or warnings for the agent to distinguish between its instructions and untrusted data.
Capability inventory: The agent can perform actions like gog gmail send and gog drive share (detailed in references/permissions.md), which could be abused if the agent follows malicious instructions found in ingested data.
Sanitization: The skill does not describe any sanitization or filtering of retrieved API data before it is processed by the agent.

google