Skip to content

Multi-Agent Spec

A portable specification format for multi-agent AI systems. Define agents, teams, and deployments once — deploy to Claude Code, Kiro CLI, AWS AgentCore, CrewAI, and more.

Why Multi-Agent Spec?

AI coding assistants and agent frameworks each have their own configuration formats. Multi-Agent Spec provides a unified way to define:

  • Agents — Individual AI agents with roles, tools, and tasks
  • Teams — Multi-agent workflows with DAG-based orchestration
  • Deployments — Platform-specific deployment configurations
  • Reports — Structured output from agent executions

Key Features

  • Write once, deploy anywhere — Define agents in a portable format, generate platform-native configs
  • DAG-based orchestration — Express complex workflows with dependencies
  • Go/No-Go semantics — Built-in status tracking inspired by NASA mission control
  • Rich reporting — Structured reports with severity, verdicts, and aggregation tags
  • Schema-validated — JSON Schema for all formats with Go and TypeScript SDKs

Supported Platforms

Platform Type Status
Claude Code AI Coding Assistant Supported
Kiro CLI AI Coding Assistant Supported
Gemini CLI AI Coding Assistant Planned
AWS AgentCore Cloud Agent Framework Supported
CrewAI Python Agent Framework Supported
AutoGen Python Agent Framework Supported
Kubernetes Container Orchestration Supported

Quick Example

Define an agent:

# agents/reviewer.md
---
name: reviewer
description: Reviews code changes for quality and security
model: sonnet
tools:
  - Read
  - Grep
  - Glob
tasks:
  - id: security-check
    description: Check for security vulnerabilities
    type: pattern
    pattern: "(password|secret|api_key)\\s*="
---

You are a code reviewer specializing in security analysis.

Define a team workflow:

{
  "name": "review-team",
  "workflow": {
    "steps": [
      {
        "name": "static-analysis",
        "agent": "reviewer",
        "depends_on": []
      },
      {
        "name": "security-audit",
        "agent": "security",
        "depends_on": ["static-analysis"]
      }
    ]
  }
}

Deploy to multiple platforms:

{
  "team": "review-team",
  "targets": [
    {
      "name": "local-claude",
      "platform": "claude-code",
      "output": ".claude/agents"
    },
    {
      "name": "local-kiro",
      "platform": "kiro-cli",
      "output": "~/.kiro/agents",
      "kiroCli": {
        "prefix": "rev_"
      }
    }
  ]
}

Getting Started

Documentation

  • Schemas — JSON Schema reference
  • CLI — Command-line interface
  • SDK — Go SDK reference

Guides