browser-harness

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The skill intentionally exposes and documents features that attach to a user's live browser sessions, sync/upload browser cookies to a cloud service, and includes examples that read authenticated page content and write it out — capabilities that can be used for credential theft and data exfiltration if misused.

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

• Third-party content exposure detected (high risk: 0.75). SKILL.md describes extracting page content from arbitrary navigated URLs via the harness’s js(...) (e.g., article.innerText / document.body.innerText), which is outsider-authored free text from the visited website and can include prompt-injection payloads that get ingested into the agent context.