The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to perform targeted code searches and network probes to verify authentication logic.
Evidence: The skill defines specific grep commands in Phase 2 to identify the use of getSession() for authorization and suggests live network probes in Phase 5 to test for IDOR and authentication bypasses.
[INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting and analyzing potentially untrusted project source code during its audit phases.
Ingestion points: Application source code files, including middleware.ts, route handlers, and server actions.
Boundary markers: The skill does not use explicit data delimiters but relies on a structured four-step reasoning framework (Observe, Interpret, Classify, Severity) to process findings.
Capability inventory: The agent is granted capabilities to perform directory-wide greps and execute network probes (e.g., via curl or similar tools) to verify route responses.
Sanitization: No explicit code sanitization or escaping mechanisms are described, though the skill includes a mandatory 'Self-critique' step to prevent the agent from making unconfirmed assumptions based on ingested code.

audit-auth-flows