continuous-learning

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

• Third-party content exposure detected (high risk: 0.70). Outsider free text can enter the LLM context because the Stop-hook script reads the full Gemini session transcript from "$GEMINI_TRANSCRIPT_PATH" (which includes user/other participants’ messages) and then triggers evaluate-session.sh for pattern extraction, i.e., it ingests session message bodies authored by parties other than the operating user.