flux-parallel-dispatch

To run independent subagents in parallel, define explicit scope, constraints, and owned files for each task. Dispatch them simultaneously when root causes differ, then synthesize results and run integration checks in the main thread.

Use parallel dispatch when tasks are completely independent with no shared state, such as split review tasks by subsystem or deploying independent scouts. Speed comes from independence, not from spawning agents aggressively.

The best way to split review tasks is by subsystem or concern, ensuring disjoint write sets. Define clear ownership and expected outputs for each unit, run them in parallel, and let one agent integrate the results later.

No, you should not coordinate parallel subagents when tasks have shared state. This approach requires cleanly separated work with disjoint write sets to guarantee safe parallel execution and successful post-execution integration.

To integrate results after parallel execution, synthesize the outputs in the main thread. The workflow guarantees integration checks are performed after all independent subagents finish their scoped tasks.

Yes, you must specify explicit scope, constraints, owned files, and expected outputs before dispatching independent scouts. Clear ownership boundaries ensure each unit runs with its own context without conflicts.