power-crawl

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The skill contains deliberate instructions to capture and reuse authentication material (localStorage tokens, cookies, intercepted GraphQL request bodies) and to perform cross-origin requests with session cookies, which are high-risk patterns for credential theft and data exfiltration despite their stated scraping purpose.

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

• Third-party content exposure detected (medium risk: 0.30). Power Crawl의 런타임은 browser_network_requests/browser_network_request/browser_evaluate를 통해 사용자(외부)가 생성/제공한 웹페이지·API 응답의 자유텍스트(응답 바디/HTML/JSON/그래프QL 응답)를 읽어 모델 컨텍스트에 섞지 않고도 파일로 저장하는 흐름이지만, 여전히 실제 LLM이 ingest하는 스키마 판별·정찰 단계에서 외부 텍스트에 노출될 수 있다(예: browser_network_request { part: "response-body" }, browser_evaluate에서 document.querySelector(...).textContent).