The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local commands such as git, ruff, npx, and bash to perform linting, history analysis, and code auditing. These commands are used for analysis and are governed by strict instructions to avoid unauthorized modifications or installations.
[INDIRECT_PROMPT_INJECTION]: As the skill's primary purpose is to review untrusted code, it naturally ingests data that could contain malicious instructions. However, the skill incorporates a 'Shared-Context Provenance Rule' and an 'Adversarial Verification Panel' with four distinct verifier lenses to independently validate findings and detect attempts to override the reviewer's judgment.
[EXTERNAL_DOWNLOADS]: The skill uses standard package managers (npx, uvx) to run analysis tools like knip and ruff from well-known public registries. It explicitly instructs agents not to install new software if these tools are unavailable, maintaining environment integrity.

review-quality-gates