optimize-agent-prompt

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow posts/patches a Browserbase agent, then reads the agent’s run output by polling GET /agents/runs/{runId}/messages and (when present) GET /sessions/{sessionId}/logs, meaning any free text produced/ingested by an outsider-authored website or page the agent visits is ingested into the teacher loop via tool-result/text parts.