The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Static string literals containing insecure command patterns, specifically curl | bash, are present in eval/run_eval.py and assets/test_audit_posture.py. These are utilized as test fixtures to verify that the scanner correctly identifies unverified remote code execution in CI/CD workflows.
[DYNAMIC_EXECUTION]: The evaluation script eval/run_eval.py dynamically creates repository structures with various configuration files and scripts, then uses the subprocess module to execute the audit_posture.py script against them for verification purposes.
[INDIRECT_PROMPT_INJECTION]: The skill's workflow requires the agent to open and adjudicate code findings from an external repository, creating a surface for indirect prompt injection via malicious comments or code.
Ingestion points: File contents from the target repository are read by assets/audit_posture.py and provided to the agent.
Boundary markers: The instructions do not define explicit delimiters or 'ignore embedded instructions' warnings for the audited content.
Capability inventory: The agent utilizes shell execution to run the auditing tool and has access to standard file system operations.
Sanitization: Repository content is read as-is and displayed without escaping or sanitization.

security-posture-audit