design-md-to-shadcn

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime workflow ingests an external DESIGN.md (optionally fetched from a URL or pasted text) and parses its markdown prose/front-matter into LLM context, which is outsider-authored free text via the “Get the DESIGN.md as a file path, pasted text, or URL (fetch raw markdown)” path.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly accepts and will fetch a user-provided DESIGN.md URL at runtime (e.g., "user-supplied DESIGN.md URL" fetched via WebFetch), and that fetched markdown directly drives the agent's mapping/editing behavior (it is parsed and used to generate prompts/instructions for edits), so an arbitrary external URL can control the agent.