xiaopu-web-design-skill

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

• Third-party content exposure detected (high risk: 0.75). The skill’s Phase A “Reference URL” path can crawl arbitrary public web pages at runtime (via scripts/crawl.py / Playwright), ingesting extracted design tokens/text from outsider-authored content into the LLM context for subsequent DESIGN.md and code generation.

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

• Potentially malicious external URL detected (high risk: 0.70). The skill's runtime crawler/analysis uses user-supplied reference URLs (e.g., https://example-portfolio.com shown in Example 2 / https://example.com used in crawl.py examples) to fetch external page content and inject extracted design tokens into the generation pipeline, so remote content can directly control the model's outputs.