subagent-orchestrator (Develop based on the Claude Code sourcemap)

Advanced Multi-Agent Coordination

The Subagent Orchestrator is a specialized skill for developers who need their AI agent to act as a project manager, not just a solo coder. It transitions your agent from a single-process tool into a coordinator capable of managing "Explorers," "Workers," and "Verifiers" via runtime-native subagent tools like spawn_agent and send_input.

What it solves

Standard LLM prompting often results in "context bloat" or agents getting stuck in loops when handling large-scale refactors or complex investigations. This skill implements a strict delegation framework that keeps the main agent focused on the user’s high-level goal while offloading bounded, parallelizable tasks to specialized subagents. It prevents "lazy follow-ups" by enforcing a Research-to-Synthesis rule, ensuring every delegation has a clear purpose and success condition.

Supported workflows

• Research: Independent codebase discovery and impact analysis.
• Implementation: Scoped coding tasks with strict ownership boundaries.
• Verification: Fresh-eye testing where a secondary agent validates the work of the primary worker.
• Preset Roles: Rapid deployment of multi-agent patterns like "Bugfix Hotpath" or "Parallel Codebase Discovery."

Why this skill?

Unlike generic prompts that just ask an AI to "be more parallel," this skill provides a rigorous operating procedure for real-time subagent management. It handles the lifecycle of worker agents (triage, handoff, synthesis, and closing) to ensure you aren't paying for idle compute or getting disjointed, unverified code back.