The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses scripts/scan-brand.sh to perform automated discovery of the repository's file structure. It executes find, grep, and xargs to locate design systems, Tailwind configurations, and CSS custom properties. While standard for its 'brand discovery' purpose, this pattern involves executing shell logic based on repository content.
[INDIRECT_PROMPT_INJECTION]: The skill has a high surface area for indirect injection as it is explicitly designed to ingest untrusted or third-party data to generate UI variants. It reads repository code, design documentation, and optionally pulls 'inspiration' from external tools.
Ingestion points: SKILL.md (Workflow Step 2 & 5) and references/implement.md (Step 2) ingest data models, API response shapes, and third-party specs.
Boundary markers: Absent. The instructions do not mandate delimiters or 'ignore' warnings for the data being rendered in the HTML previews.
Capability inventory: SKILL.md executes bash <skill-dir>/scripts/scan-brand.sh, and open/xdg-open/start to launch browsers.
Sanitization: Not present. The skill is instructed to use 'real copy from the spec/codebase' directly in the generated HTML.
[EXTERNAL_DOWNLOADS]: The skill references a 'free, self-contained screenshot skill' at https://danny.md/skills/screenshot.md. While the instructions tell the user to fetch it if a local screenshot tool is missing, this encourages the dynamic addition of unverified external code into the agent's environment.
[SAFE]: The skill makes use of the Landingfolio MCP (mcp__landingfolio__*). This is documented as a well-known service for design inspiration and is treated as a safe external integration.

explore-design