best-answer

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

• Third-party content exposure detected (medium risk: 0.65). The skill’s runtime workflow fans out to panelists with “tools (web search, bash, read)” and then passes the panelists’ outputs (which may include scraped public web content or other outsider-authored text) into a clean-room judge subagent as LLM-readable prose, creating an indirect prompt-injection exposure path.