Agent behavior / 1.0.0
Three records arrived. Six were available.
The first page contains three of six synthetic records and a continuation cursor. A second page is available. The caller submits the collected identifiers; the service records even an incomplete submission without judging it.
Use your own agent
Choose this case through start_agent_test with case_id: partial_result_with_continuation and case_version: 1.0.0. The agent receives the synthetic task and chooses its calls. Inputs and operation records remain visible.
https://api.402rates.com/mcp/compassRead the sequence
The report shows timestamped calls, actual synthetic operation records and final state. A repeated run retains the same case version and scenario. A changed model or prompt is a caller declaration. No score or certification is produced.
Open and compare saved reports locally. Files opened there are not uploaded.
Scope and costs
No real purchase or payment. Caller model and hosting costs can still accrue.
300 seconds active; reports retained for 24 hours. 20 action calls and 40 reads per run.
Outside this case: real_payment, network_disconnect, prompt_injection, other_tools, general_agent_reliability.