skill-comply

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

• Third-party content exposure detected (medium risk: 0.30). The classifier step in scripts/classifier.py builds a prompt containing the session tool-call event.input/event.output fields (from scripts/runner.py::_parse_stream_json) and sends them to claude -p, meaning outsider-authored free text inside tool inputs/outputs is ingested for LLM classification.