The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill involves processing untrusted data sources, such as User-Agent strings and external API responses, during its 'Reachability Gate' and 'Measurement' workflows.
Ingestion points: SKILL.md (Workflow Step 2) and references/agent-traffic-measurement.md (Detection sources).
Boundary markers: The skill does not use explicit delimiters but includes strong warnings to the user to treat external content as untrusted and potential injection paths.
Capability inventory: The agent is instructed to make network requests to verify surface reachability.
Sanitization: The skill advises implementing machine-readable errors and sanitization in the platform being designed.
[SAFE]: The skill is educational and focuses on best practices for security and platform engineering. It explicitly mitigates risks by recommending short-lived scoped credentials, out-of-band human approvals for destructive actions, and robust bot verification methods.

agent-integration-surfaces