skill-idea-miner

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

• Third-party content exposure detected (high risk: 0.80). The skill reads and ingests user-generated Claude session logs from ~/.claude/projects/ (see SKILL.md and scripts/mine_session_logs.py: parse_session → collect user_messages and pass user_samples into abstract_with_llm/_build_llm_prompt), and those untrusted messages are used by the LLM to generate candidates and influence downstream actions, so third‑party content can materially affect behavior.