deep-research

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime retrieval phase uses search-cli/WebSearch to fetch and extract public web pages (outsider-authored content) and then feeds extracted prose into the LLM for synthesis/packaging (e.g., Phase 3 RETRIEVE → Phase 5 SYNTHESIZE), creating an indirect prompt-injection risk.