continuous-learning-v2

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

• Third-party content exposure detected (medium risk: 0.30). The required runtime workflow runs an observer that reads Gemini tool hook data written into ~/.gemini/homunculus/observations.jsonl and then sends that free-text (including tool inputs/outputs from the user’s sessions) to the Haiku model for pattern detection.