audit-noise

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

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md via the required runtime workflow (run scripts/detect.sh), the agent reads arbitrary file contents from uncommitted/queued repository markdown targets (git status ... | awk '/\.md$/ {print $NF}' or provided --paths-file), and LLM logic ingests those lines to classify noise shapes.