website-to-hyperframes

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

• Third-party content exposure detected (high risk: 0.95). Step 1 runs npx hyperframes capture <URL> and then reads capture/extracted/visible-text.txt (outsider-authored page text from the user-supplied URL) into the agent’s working summaries, which become LLM context for later steps.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly runs a runtime capture of arbitrary websites via "npx hyperframes capture " (example: https://stripe.com), then reads the fetched capture files (tokens.json, shaders.json, asset descriptions, visible-text, etc.) which directly drive DESIGN.md, SCRIPT.md, STORYBOARD.md and composition code—meaning user-supplied external URLs fetched at runtime can inject content that controls prompts and executable assets (shaders/code) used by the agent.