mma-delegate

To delegate ad-hoc tasks, dispatch a flat prompt with optional target paths and acceptance criteria to a worker. This offloads execution to cheaper workers, returning a structured per-task report while preserving your primary model's main-context tokens for higher-value reasoning.

Task dispatching sends a single ad-hoc task to a worker when no plan file exists on disk. The endpoint accepts a flat prompt and context blocks, returning a taskId with a pollable status to track the worker's progress and retrieve the structured report.

Yes, you can track delegated task status using the pollable status endpoint. When you dispatch a task with a flat prompt and context blocks, the system returns a taskId that you use to poll the endpoint and monitor the worker's execution progress until completion.

Yes, mma-delegate is specifically designed for situations when mma is running but no plan file exists on disk. It allows you to choose this dispatching method to save main-context tokens while delegating quick research or small refactors to cheaper workers.

The best way to handle small refactors without exhausting context is delegating them as ad-hoc tasks to cheaper workers. By supplying a flat prompt and optional scope, workers execute the refactors and return a concise structured report, keeping the main agent lean.

You should avoid ad-hoc task delegation when a plan file already exists on disk for your workflow. The dispatching mechanism is specifically built for on-demand tasks without prebuilt plans, so utilizing it alongside existing plan files may not yield the intended context-saving benefits.