llm-council-harness

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

• Third-party content exposure detected (high risk: 0.75). The required workflow ingests outsider-authored free text from each CLI subprocess’s stdout (including potentially adversarial, JSON-looking content) into the agent’s LLM context via the boss prompt assembly (anon_block/boss_prompt) and subsequent _complete call, creating an indirect prompt-injection risk.