multi-agent-orchestration

Overview

Multi-agent orchestration is a meta-skill for coordinating multiple AI coding agents in parallel, pipeline, or swarm modes. It augments the orchestrator (primary session agent) with a six-step workflow: Detect parallelizability, Decompose into non-overlapping sub-tasks, Delegate via the task tool with scoped context, Collect and verify each subagent's output independently, Integrate merged files, and Verify with a full build and test suite.

Use this skill when a task requires more than two agents, involves multi-module refactors with independent files, or follows a spec-driven pipeline with phase gates. Subagents must never edit overlapping files, must receive only their relevant skill (not the full AGENTS.md), and must never commit — only the orchestrator commits after verification.

The output is merged, tested source code from coordinated subagents. The skill integrates with incremental-implementation (build), test-driven-development (tests), and code-review-and-quality (review) as subagent roles within the pipeline.