balise-visual-references

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

• Third-party content exposure detected (medium risk: 0.30). The required runtime workflow can ingest outsider-authored free text indirectly by browsing and inspecting “exact source URLs” from a user-driven visual research task (e.g., gallery pages that contain arbitrary user-generated copy), via the agent’s “Conduct direct research” step.