dialectic

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

• Third-party content exposure detected (high risk: 0.85). At runtime, Phase 1’s “research” and Phase 4’s “donor recruitment” can fetch public web content (e.g., Wikipedia/random-article donors and “web_search, web_fetch” usage), whose extracted page text is then written into the context briefing file read by subagents and therefore becomes readable free text inside the agents’ LLM prompts.