The Agent Skills Directory

[SAFE]: The skill implements a structured multi-phase workflow (Detect, Load Context, Evaluate, Report, Fix) that ensures systematic analysis rather than reactive behavior.
[SAFE]: Data access is limited to reading the targeted codebase for audit purposes and scanning sibling skill directories to load relevant local rules for detected technologies.
[SAFE]: The skill explicitly instructs the agent to search for security vulnerabilities, including hardcoded secrets and injection patterns, acting as a defensive security tool.
[SAFE]: The 'Audit and fix' capability is protected by a mandatory human-in-the-loop control, requiring the agent to present findings and receive explicit user authorization before any file modifications are performed.
[PROMPT_INJECTION]: The skill processes untrusted data from the codebase being audited, creating a surface for indirect prompt injection. However, this is mitigated by strict workflow enforcement and the requirement for user approval before taking action.
Ingestion points: Target codebase source files and dependency manifests; sibling skill configuration files.
Boundary markers: The skill uses structured report segments and headers to separate analysis from output.
Capability inventory: MCP tool execution (resolve-library-id, query-docs), file reading across the codebase and sibling skills, and file writing (for approved fixes).
Sanitization: All automated fixes require prior human review and explicit confirmation of the audit report.

code-audit