03-performance-eval-global

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

• Third-party content exposure detected (high risk: 0.80). The SKILL.md explicitly instructs the agent to "Auto-pull data" from external MCP integrations (Meta Official MCP, Google Ads MCP, TikTok Ads MCP, Pipeboard) in Step 0 and uses those pulled metrics in the diagnostic flow and decision tree, which means the agent ingests third‑party ad/creative data that can materially influence its actions.