The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches UI components using the shadcn/ui CLI and installs additional agent skills from Vercel Labs' public repository to ensure code quality.
[COMMAND_EXECUTION]: Executes shell commands to scaffold projects, install dependencies, and run build scripts. Examples include npx shadcn@latest add, npm install, and npm run build.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection through the processing of untrusted markdown files.
Ingestion points: Markdown research reports provided by the user (parsed in SKILL.md).
Boundary markers: Supports explicit HTML comment markers like  to guide parsing, but lacks explicit instructions to ignore potentially malicious embedded prompts within the report text.
Capability inventory: The skill writes React components, Vite configuration files, and data structures to the file system, and executes package installation and build commands.
Sanitization: Content is rendered via react-markdown with remark-gfm, but the mapping logic that determines component composition is driven by parsing the structure of the input document.
[DYNAMIC_EXECUTION]: Generates executable React code and project configuration files (e.g., vite.config.js, reportData.js) at runtime based on the detected structure and markers in the input markdown documents.

site-generation