adversarial-elixir

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

• Third-party content exposure detected (high risk: 0.70). The required runtime workflow composes a reviewer prompt that includes the review target content/diff (outsider-authored code or PR text if the operating user didn’t author it), which is then sent verbatim to the blind reviewer Task subagent as LLM context—so outsider free text can be ingested via the “Dispatch one blind reviewer” path.