The Agent Skills Directory

[COMMAND_EXECUTION]: The skill incorporates browser automation for UI verification, directing the agent to use local tools like Playwright and Chrome DevTools. These tools are leveraged to capture and compare UI states during the implementation process within the local environment.- [DATA_EXPOSURE]: The skill manages project data by persisting concept decisions in localized temporary directories. To ensure project security and prevent accidental exposure in public repositories, the instructions mandate adding these temporary paths to the .git/info/exclude file, demonstrating a strong focus on data hygiene and repository integrity.- [PROMPT_INJECTION]: The skill identifies an indirect prompt injection surface when processing external reference websites. This is addressed through a structured security posture: (1) Ingestion points: Browser-captured content and reference URLs provided by the user; (2) Boundary markers: Explicit instructions to treat external sites strictly as visual and interaction references rather than authoritative instruction sources; (3) Capability inventory: Access to local files, browser automation (Playwright), and local file system writes; (4) Sanitization: Mandatory guidelines to avoid copying tracking scripts, analytics, or proprietary implementation details from external sites.

ui-ux-concept-implementation