web-clone

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

• Third-party content exposure detected (medium risk: 0.65). Required runtime workflow includes fetching and ingesting outsider-authored web page text/HTML from arbitrary user-supplied URLs via Playwright-based probes (e.g., scripts/recon-site.mjs and scripts/interaction-probe.mjs call page.evaluate to extract document.title, headings/body text, and document.documentElement.outerHTML into JSON/markdown). That readable free text is then used by the agent context in subsequent reasoning/outputs, creating indirect prompt-injection exposure.