agent-orchestration

Orchestrate agent-driven task decomposition and coordination for complex coding projects.

1|Updated Mar 18, 2026
One-click install
npx skills add https://github.com/xxih/ai-harness-zh --skill agent-orchestration-xxih
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: agent-orchestration
Source: https://github.com/xxih/ai-harness-zh/tree/main/packages/agent-orchestration/targets/codex/skills/agent-orchestration
Command: npx skills add https://github.com/xxih/ai-harness-zh --skill agent-orchestration-xxih

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill helps a main agent keep a global view while coordinating multiple subagents in complex coding efforts, preventing duplicated work, vague delegations, missed verification, and chaotic merges across modules.

Core Features & Use Cases

  • Role layering and phased workflow: defines planner, orchestrator, worker, explore, and reviewer roles and recommends a default phase order of research -> plan -> execute -> review -> verify.
  • Single-task delegation templates: creates low-ambiguity, verifiable prompts that specify inputs, non-goals, expected outputs, and blocking conditions.
  • Parallelism independence checks and anti-duplication rules: evaluates write-set overlap, dependency order, and merge costs before allowing parallel execution.
  • Result collection and verification gates: requires the main agent to read diffs, code, tests or evidence and decide accept / follow-up / re-dispatch.
  • Use cases include multi-module refactors, cross-cutting bug fixes that need separate research and implementation phases, and orchestrating multi-agent experiments where verification and non-overlap are critical.

Quick Start

Ask the agent to produce an orchestration note, split the work into verifiable single-task delegation prompts, run independence checks before any parallel work, and list verification gates for each subtask.

Frequently Asked Questions about agent-orchestration

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I coordinate multiple AI agents for a complex coding project?

Multi-agent coding coordination requires an orchestrator to decompose tasks into planner, worker, and reviewer roles. You can manage this by generating single-task delegation prompts with specific inputs, non-goals, and verification criteria to prevent duplicated work and chaotic merges.

What is the best way to prevent subagents from overwriting each other's code during parallel execution?

To prevent subagents overwriting code during parallel execution, perform parallelism independence checks. This involves evaluating write-set overlap, dependency order, and merge costs before dispatching tasks to ensure safe, non-overlapping multi-agent workflows.

How do I create verifiable delegation prompts for AI coding assistants?

Verifiable delegation prompts are created using templates that specify inputs, expected outputs, non-goals, and blocking conditions. This low-ambiguity format allows the main agent to read resulting diffs and tests to decide whether to accept, follow up, or re-dispatch the task.

When do I need multi-agent orchestration for code refactoring?

You need multi-agent orchestration for cross-module refactors, multi-step bug fixes, and research-to-implementation workflows. It is necessary when a main agent must maintain a global view while coordinating separate research, plan, execute, review, and verify phases across subagents.

Can I use task decomposition for cross-cutting bug fixes that require separate research and implementation phases?

Yes, task decomposition supports cross-cutting bug fixes by layering roles into planner, orchestrator, worker, and reviewer. It structures the workflow into research, plan, execute, review, and verify phases, producing orchestration records and verification gates for each subtask.

How do I verify the results of delegated coding tasks from AI subagents?

Verifying delegated coding tasks requires setting up verification gates where the main agent reads diffs, code, and test evidence. The agent then decides to accept, follow up, or re-dispatch the work based on the expected outputs and blocking conditions defined in the delegation prompt.