analyze-eval

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

• Third-party content exposure detected (high risk: 0.75). SKILL.md instructs querying an internal action (debug:getEvalDebugInfo) using a user-supplied evalId, and the returned JSON includes evalSourceFiles/task text/failureReason/file contents that are outsider-authored (from the eval’s stored source and task), which would be ingested into the agent’s LLM context for analysis.