The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill can download and execute the well-known react-doctor utility from npm for React and Next.js projects. This operation is gated by an explicit user confirmation prompt to ensure transparency and authorization.
[COMMAND_EXECUTION]: Employs standard shell commands including find, grep, and git to perform mechanical auditing tasks such as identifying project structure, detecting dependencies, and scanning for common code quality issues.
[DYNAMIC_EXECUTION]: Executes a vendor-provided Python script (scripts/codebase-map.py) to generate a structural map of the repository as part of the initial detection phase.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes arbitrary code from the user's repository, creating a potential surface for instructions embedded in data to influence reviewer agents.
Ingestion points: Reads repository files using Glob and Grep tools during detection and provides file content to specialized reviewer agents.
Boundary markers: Requires user confirmation of the detected project stage and mandates that agents verify findings against the actual source code before reporting.
Capability inventory: Performs shell command execution (grep, find, npx), script execution, and file writing to the docs/arc/audits/ directory.
Sanitization: Employs high-capability models for reasoning and includes a dedicated findings vetting phase (Phase 4) where high-severity leads are cross-referenced with source files to filter out false positives or adversarial content.

audit