Can agents communicate with each other?

Yes, when using 'Agent Teams' mode, workers can discuss findings, challenge each other, and coordinate across different layers of the stack.

Does the orchestrator resolve conflicts between agents?

Yes, one of the core functions of the 'delegate' skill is Curation, where Claude validates findings, filters false positives, and resolves conflicting tool outputs into a final recommendation.

How does it handle task dependencies?

The skill uses a Directed Acyclic Graph (DAG) pattern to decompose features into atomic tasks, declaring dependencies so that subsequent phases only execute once prerequisites are completed.

When should I use DAG orchestration versus simple delegation?

Simple parallel spawning is best for features involving 1-5 files. DAG orchestration is recommended for large migrations or multi-feature releases involving 10+ subtasks and multiple phases.

What models can this skill delegate to?

It primarily uses the Moonbridge MCP to access Codex and Kimi adapters, but also integrates with Gemini CLI and Thinktank for multi-model consensus and research.

AI Multi-Agent Orchestrator

Name: AI Multi-Agent Orchestrator
Author: phrazzld

byphrazzld

•

생산성 및 워크플로

Orchestrates specialized AI agents to perform complex analysis, code reviews, and architectural audits through parallel delegation and synthesis.

This skill transforms Claude into a high-level project lead capable of delegating complex technical tasks to a specialized team of AI models, including Codex, Kimi, and Gemini. By implementing a standardized 'Route-Collect-Curate-Synthesize' pattern, it enables the execution of high-effort operations such as deep security audits, large-scale refactors, and multi-perspective investigations. It features a robust DAG-based orchestration engine for managing dependencies in multi-phase migrations, ensuring that complex workflows are executed efficiently with fresh context for every sub-agent.

주요 기능

01Parallel agent execution to significantly reduce processing time for large tasks

02Multi-backend orchestration using Moonbridge MCP for Codex and Kimi models

03Automated synthesis of conflicting AI perspectives into unified action plans

04Integration with specialized domain agents for security, performance, and architecture

05DAG-based task scheduling for complex, multi-phase migrations and releases

065 GitHub stars

사용 사례

01Performing comprehensive security and performance audits across large repositories

02Executing multi-phase legacy code migrations involving backend, frontend, and testing

03Orchestrating a multi-model 'think tank' to validate complex architectural decisions

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add phrazzld/claude-config delegate

For use in Claude.ai and ChatGPT

Download Skill