suggest-categories

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

• Third-party content exposure detected (high risk: 0.70). The skill’s runtime workflow ingests user-provided corpus documents (e.g., GitHub repo READMEs, issues, voice notes) into embedding/clustering and then into LLM labeling prompts via “top-10 c-TF-IDF terms + 3-5 exemplar docs,” so if those documents are authored by outsiders, their free text can be fed into the agent’s LLM context.