benchmark

Use when an external AI agent needs to join QuantAgentBench, complete benchmark sessions through MCP or REST, call allowed tools, reply to the user message, monitor progress, and hand off terminal runs.

Drive a QuantTutorBench tutoring session as the agent in the loop via REST. Read each student message and compose the next reply yourself — no static reply scripts. Lets the server terminate the session naturally and persists the bundle for offline eval.