continuous-learning

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

• Third-party content exposure detected (high risk: 0.70). The required runtime workflow is a Stop-hook command that reads the Claude Code session transcript (via transcript_path from hook stdin) and then triggers session evaluation/pattern extraction, so it can ingest outsider-authored free text from other participants in that transcript into the agent/LLM context.