last-30-days-research

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

• Third-party content exposure detected (high risk: 0.95). The skill’s runtime workflow performs web searches and then reads Reddit/X thread titles and comments (outsider-authored free text) into the agent’s LLM context via the “read the thread…top-level comments…highly-upvoted replies” and “Note: X search…supplement with searches…” steps.