page-reduce

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

• Third-party content exposure detected (high risk: 0.85). Yes—Step 3 opens a user-supplied URL, runs Phase 1 in the browser to extract tokenizedHtml from outsider-authored webpage DOM text, then Step 3 returns that JSON via playwright-cli eval into the agent’s LLM context for Phase 2 reasoning.