The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute standard security auditing tools via the command line, specifically semgrep with various configuration profiles (--config auto, --config p/owasp-top-ten). This behavior is aligned with the skill's stated purpose for white-box auditing.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill reads local project context from ../field-journal/precedent-pentest.md to establish audit authorization and scope. While this involves accessing files outside the immediate skill directory, it is a functional requirement for localized security reviews and does not involve exfiltration to external domains.
[INDIRECT_PROMPT_INJECTION]: The skill has an inherent attack surface for indirect prompt injection as it is designed to ingest and analyze untrusted source code and third-party security reports.
Ingestion points: Untrusted source code repositories and project journals (../field-journal/precedent-pentest.md).
Boundary markers: The instructions do not define explicit delimiters for untrusted content, though they emphasize manual verification of all findings.
Capability inventory: Execution of SAST tools (Semgrep, CodeQL, Bandit, etc.) through the shell.
Sanitization: No specific sanitization or escaping of the ingested code is described, relying instead on the parsing logic of the auditing tools and human-in-the-loop verification.

code-audit