The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external websites via Firecrawl to generate code, creating an attack surface where malicious content on a target site could influence the agent's code generation.
Ingestion points: External URLs are scraped using the firecrawl-mcp___firecrawl_scrape tool as defined in Phase 1 of SKILL.md.
Boundary markers: The skill implements a mandatory 'Phase 2: Analysis' where a report based on references/analysis-template.md must be approved by the user before code is generated. While this provides a human-in-the-loop checkpoint, there are no explicit instructions for the agent to ignore hidden commands within the scraped HTML/Markdown.
Capability inventory: The skill can create a full Next.js project structure, write multiple code files (app/page.tsx, components/landing/*.tsx), and perform network requests to download images via fetch as specified in the Image Handling section of SKILL.md.
Sanitization: There is no explicit sanitization or filtering of the scraped content mentioned in the instructions before it is used for component generation.
[EXTERNAL_DOWNLOADS]: The skill fetches images from arbitrary URLs discovered during the scraping process. It also uses Unsplash as a fallback service, which is a well-known and reputable image provider.
[COMMAND_EXECUTION]: The documentation in references/tech-stack.md references the use of npx shadcn@latest add to install UI components. These are standard commands for the referenced technology stack and represent intended developer workflow rather than malicious execution.

clone-website