health

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

• Third-party content exposure detected (high risk: 0.85). Runtime path: scripts/collect-data.sh reads prior conversation .jsonl files from ~/.claude/projects/-${PROJECT_KEY}/... and, in deep mode, extracts their free-text messages via jq (extract_messages_from_file / print_conversation_extract) and prints them into the collector output that is then fed into the agent’s LLM context.