dispatching-parallel-agents

To dispatch subagents safely, use an independence gate to verify tasks have no shared state, then generate per-domain agent prompts and a merge report to guide integration. This maximizes throughput while preventing merge conflicts across independent code, tests, or data tasks.

You should stay in the main context instead of dispatching parallel subagents when the independence gate detects shared state or dependencies between tasks. If tasks cannot run without overlapping state, parallel dispatch risks conflicts and merge disasters.

An independence gate is a verification mechanism that checks whether two or more tasks can run in parallel without shared state. It ensures safety and isolation before dispatching subagents, preventing merge conflicts and ensuring deterministic execution across independent domains.

You can assign different prompts by using per-domain prompt orchestration, which generates specific agent prompts for each independent problem domain. This keeps work isolated and deterministic, returning a dispatch decision and merge report for safe integration.

Yes, you can use parallel subagent dispatch for test triage and data gathering simultaneously. The skill applies to scenarios with two or more independent tasks across code, tests, or data, verifying independence before parallelization to ensure safe execution.

When dispatching parallel subagents, you should expect three main outputs: a dispatch_decision indicating whether to parallelize, agent_prompts containing per-domain instructions, and a merge_report supporting safe integration of the completed independent tasks.