The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions (Gate 4) direct the agent to execute numerous standard security scanning tools such as npm audit, semgrep, bandit, gitleaks, and trivy to gather evidence for the audit. These operations are conducted locally on the project files within the approved scope.
[DYNAMIC_EXECUTION]: The skill uses a local Python script (scripts/validate_report.py) in Gate 7 to perform structural and content validation of the generated SECURITY-AUDIT.md report. The script utilizes standard libraries for regex-based parsing and does not perform network operations.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted code and configuration files from the project being audited, which creates an inherent attack surface for indirect prompt injection.
Ingestion points: Project source code, manifests (e.g., package.json), documentation (README.md), and CI/CD configurations inspected during the reconnaissance and manual review phases.
Boundary markers: The skill emphasizes a defensive posture and adherence to a defined scope, but does not provide specific delimiters or instructions for the agent to disregard natural language commands embedded within the audited project data.
Capability inventory: The agent can execute shell commands for scanners, run the bundled validation script, and modify project files if the user requests remediation.
Sanitization: The skill includes mandatory rules for redacting secrets, personal data, and customer data from the final report artifact using [REDACTED] placeholders.

security-audit