The Agent Skills Directory

[PROMPT_INJECTION]: The skill documentation includes specific instructions for the agent to avoid treating data from external API responses, error messages, or documentation as instructions. This serves as a proactive defense against indirect prompt injection where untrusted external content might attempt to manipulate the agent's behavior.
[DATA_EXFILTRATION]: The skill facilitates authenticated network requests to well-known service providers. Security is maintained through server-side logic that injects credentials and restricts requests to specific provider hosts, effectively preventing the agent from leaking local environment secrets or accessing unauthorized internal URLs.
[COMMAND_EXECUTION]: The query-staged-dataset action allows for data transformation using in-process TypeScript. While this involves dynamic execution, it is scoped to the manipulation of data staged in the platform's scratch datasets.
[EXTERNAL_DOWNLOADS]: The skill references various well-known services (e.g., Stripe, Slack, HubSpot, BigQuery) for data retrieval. All mentioned resources are official provider endpoints or documentation sites. The skill correctly advises against installing unverified or unlicensed third-party plugins.

provider-api