The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to perform network requests using curl against external targets to analyze server responses. This exposes the agent to processing potentially malicious instructions returned by an attacker-controlled endpoint.
Ingestion points: SKILL.md (instructions for executing curl against target endpoints).
Boundary markers: None identified. The skill does not provide specific delimiters or instructions to ignore content within the HTTP response bodies.
Capability inventory: The skill utilizes curl for network operations and advises on using findings for other tools like sqlmap or Burp Suite.
Sanitization: None. The agent is encouraged to read the response as a "signal," which requires interpreting the untrusted content directly.

tools-curl