The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local security binaries including bin/brakeman, bin/bundler-audit, and rspec to perform vulnerability scans and policy verification.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project source code and security scanner outputs which represents an indirect injection surface.
Ingestion points: Source files within app/ and config/ directories (Controllers, Models, Views, etc.) and tool output from Brakeman and Bundler Audit.
Boundary markers: Absent; the skill does not use delimiters to isolate untrusted code from auditing instructions.
Capability inventory: Execution of local shell commands via the agent's shell environment.
Sanitization: The skill does not implement sanitization for the code content it evaluates.

security-audit