icp-definer

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

• Third-party content exposure detected (high risk: 0.95). SKILL.md directs the agent to perform “extensive web research” and to “capture at least 3-5 real quotes or posts” from outsider-authored sources (e.g., Reddit/HN/Twitter/forums/blogs review sites) and then synthesize them into the LLM-visible market research summary and evidence used for subsequent prompts.