continuous-learning

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

• Third-party content exposure detected (high risk: 0.75). The Stop hook runs evaluate-session.sh, which reads the Claude Code session transcript from transcript_path (a file containing the full session transcript, including any outsider-authored free text the user did not choose to introduce) and then passes that transcript onward for pattern extraction into the LLM context.