The Agent Skills Directory

[SAFE]: The skill is entirely composed of documentation and knowledge base files (Markdown and JSON). It does not contain any executable scripts (.js, .py, .sh) that run on the host during analysis.\n- [DATA_EXPOSURE]: The rule examples (e.g., sec-client-data-exposure.md, qual-no-hardcoded-secrets.md) contain placeholder credentials such as sk_live_abc123 and sensitive file paths like ~/.env.local. These are provided solely as educational examples of insecure practices and do not represent actual secrets or malicious access.\n- [COMMAND_EXECUTION]: The documentation includes 'Detection hints' that suggest using command-line tools like grep and find to scan codebases. These are benign instructions intended for the developer or as logic for the agent to apply when performing a review.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted code (e.g., pull requests or source files) during its primary function (code review). While this creates an indirect injection surface, the skill provides structured rules to guide the agent, and no patterns were found that would cause the agent to execute untrusted code or ignore safety guidelines.

beforemerge-nextjs-review