The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (MEDIUM): The skill instructs agents and users to use npx shadcn@latest add [URL] to install components directly from remote sources like 21st.dev or custom Vercel deployments (found in rules/marketplaces.md and rules/registry.md). This pattern facilitates the execution of remote code from untrusted domains. The severity is downgraded from HIGH to MEDIUM as this is a standard practice for the component distribution primary purpose of this skill.
[COMMAND_EXECUTION] (MEDIUM): Multiple files including README.md and rules/npm.md contain shell commands for file manipulation (cp -r), package management (npm install, npm publish), and deployment (vercel --prod). An agent following these instructions might execute these commands if granted terminal access.
[PROMPT_INJECTION] (LOW): The skill is vulnerable to Indirect Prompt Injection (Category 8) because it is designed to analyze and refactor user-provided code.
Ingestion points: Processes user-provided React source code and component specifications as described in SKILL.md.
Boundary markers: Absent. There are no instructions to ignore malicious directives embedded in comments or documentation within the processed code.
Capability inventory: The skill possesses code generation and modification capabilities across all rule files.
Sanitization: None. The skill does not implement validation or escaping for the external content it interpolates into its logic.
[EXTERNAL_DOWNLOADS] (LOW): The skill references several non-whitelisted domains for documentation and component fetching, including components.build, 21st.dev, and radix-ui.com. While these are contextually relevant, they represent untrusted external sources.

components-build