skill-idea-miner

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text from Claude Code session logs is ingested: parse_session() reads ~/.claude/projects/*/*.jsonl and extracts userType=="external" user message content into user_messages, which is then embedded into the runtime prompt sent to claude -p via _build_llm_prompt() → abstract_with_llm().