paper-to-code

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

• Third-party content exposure detected (high risk: 0.85). The skill ingests outsider-authored free text from the paper itself (e.g., algorithm box/equations/prose) at runtime to extract the method and assumptions, which is outsider content fed into the agent’s LLM context for implementation planning.