The Agent Skills Directory

[COMMAND_EXECUTION]: The script eval/run_eval.py uses subprocess.run to invoke the skill's main auditing script, assets/audit_posture.py, for automated testing. This execution is limited to internal diagnostic tools and synthetic test repositories.
[EXTERNAL_DOWNLOADS]: The eval/run_eval.py and assets/test_audit_posture.py files contain strings representing curl | bash commands. These are static test cases designed to confirm the scanner's detection capabilities and are never executed.
[PRIVILEGE_ESCALATION]: Testing scripts programmatically apply the setuid bit to local temporary files to ensure the audit tool can detect binaries with elevated privileges. This occurs only within the controlled test environment.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the repository being audited. A malicious repository could contain instructions designed to deceive the agent during the manual adjudication step of the audit process.
Ingestion points: Repository files read by assets/audit_posture.py and the agent's workspace.
Boundary markers: None; the skill relies on deterministic regex matching.
Capability inventory: eval/run_eval.py (subprocess), assets/audit_posture.py (file system read).
Sanitization: Not applicable; the tool performs regex-based scanning of raw file content.

security-posture-audit