multi-agent-runner

You can compare results by running a manually-invoked workflow where multiple agents execute the same task simultaneously. The system collects their outputs side-by-side, helping you evaluate and choose the best outcome.

Multi-agent execution supports general (non-editing) mode for analysis and planning tasks, and code (editing) mode for direct file modifications. Both modes let you specify custom models, strategies, and constraints per agent.

Yes, multi-agent workflows fully support user-specified models, strategies, and constraints. You can configure each agent independently to test different approaches and compare which combination yields the best result.

Use a multi-agent runner for code optimization when you need to evaluate multiple approaches simultaneously. Running several agents on the same optimization task lets you compare strategies and select the most effective outcome rather than relying on a single attempt.

The multi-agent runner requires explicit manual invocation by the user. It does not trigger automatically; you must intentionally start the workflow when you want to compare results from multiple agents on a specific task.