The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (SAFE): The skill downloads images from Unsplash and uses Firecrawl to scrape website data. These are known services and the skill uses them according to its stated purpose.
[COMMAND_EXECUTION] (LOW): The skill suggests running npx shadcn@latest add ... to install UI components. While standard for Next.js development, installing packages based on external scraping results is a minor risk.
[INDIRECT PROMPT INJECTION] (LOW): The skill's primary function is to ingest data from an attacker-controlled website (via Firecrawl) and interpolate that data into code generation prompts.
Ingestion points: firecrawl-mcp___firecrawl_scrape in SKILL.md phase 1.
Boundary markers: Present. The skill uses a mandatory Phase 2 analysis step where the user must review and confirm the extracted data before code generation occurs.
Capability inventory: File writing (app/, components/, public/images/) and package installation (npx shadcn).
Sanitization: Not explicitly defined in instructions beyond the 'Analysis Template' review by the user. An attacker could embed malicious instructions in the HTML of a site meant to be 'cloned' to influence the generator's behavior.

clone-website