Team Copilot: Multi-Agent AI Collaboration System

Coordinate multiple specialized AI agents to solve complex tasks through collaboration.

Features

✨ APEV Workflow - Ask → Plan → Execute → Validate with auto-retry
🔄 YAML Pipelines - Define custom agent workflows with DAG execution
🤖 4 Specialized Agents - Architect, Implementer, Reviewer, Analyst
🔍 Built-in Validation - Every output is reviewed before delivery
⚡ Fault Isolation - Each agent runs independently for robustness

Quick Start

Prerequisites

• Node.js 18+
• GitHub Copilot CLI (install guide)

Installation

npm install -g multi-agent-copilot-sdk

Usage

APEV Mode (Interactive collaboration):

copilot-agents

Pipeline Mode (YAML-defined workflows):

copilot-pipeline examples/simple-pipeline.yml --prompt "Your task"

Try It Out

APEV Mode Examples

# Interactive session
copilot-agents
> Create a TypeScript calculator with add, subtract, multiply, divide

# Or in development:
npm run ape

Agents collaborate: Bob plans → Alice implements → Dave validates → Auto-retry if needed

Pipeline Mode Examples

Simple 3-agent pipeline:

# examples/simple-pipeline.yml
name: "Code Review Pipeline"
agents:
  - id: implementer
    profile: IMPLEMENTER
  - id: reviewer
    profile: REVIEWER
    dependsOn: [implementer]
  - id: verifier
    profile: ANALYST
    dependsOn: [reviewer]

Run it:

# After global install:
copilot-pipeline examples/simple-pipeline.yml --prompt "Build a REST API"

# In development:
npm run pipeline examples/simple-pipeline.yml -- --prompt "Build a REST API"

Agent Profiles

| Agent | Role | Expertise | |-------|------|-----------| | Alice | Implementer | Code generation, file creation | | Bob | Architect | System design, planning | | Carol | Analyst | Analysis, debugging, insights | | Dave | Reviewer | Code review, validation |

Pipeline Configuration

Create custom workflows in YAML:

name: "My Pipeline"
version: "1.0"

agents:
  - id: "step1"
    name: "Data Processor"
    profile: "IMPLEMENTER"
    customInstructions: "Process CSV data"
  
  - id: "step2"
    name: "Quality Check"
    profile: "REVIEWER"
    dependsOn: ["step1"]

settings:
  continueOnFailure: false  # Stop on first error
  maxParallelAgents: 3      # Limit concurrency
  verbose: true

See Pipeline Guide for complete reference.

Architecture

Per-Agent Isolation:

• Each agent runs its own CopilotClient process
• Independent stdio streams eliminate contention
• Failures in one agent don't cascade to others

DAG Execution:

• Topological sort determines execution order
• Parallel execution of independent agents
• Dependency validation prevents cycles

Coordinated Cleanup:

• Promise.allSettled() for parallel shutdown
• Proper error logging (no suppression)
• 100ms stream flush delay for graceful termination

Details: Architecture Docs | Implementation Summary

Development

# Clone and install
git clone https://github.com/tdevere/Team-Copilot.git
cd Team-Copilot
npm install

# Build
npm run build

# Test APEV mode
npm run apev

# Test pipeline
npm run pipeline examples/simple-pipeline.yml -- --prompt "Your task"

# Run tests
npm test

Global Installation:

npm install -g .
# Then use copilot-agents and copilot-pipeline directly

Configuration (Optional)

# Custom Copilot CLI path
export COPILOT_CLI_PATH=/path/to/copilot

# Debug logging
export COPILOT_LOG_LEVEL=debug

Troubleshooting

"Cannot find copilot CLI"

• Install: winget install GitHub.Copilot (Windows) or brew install copilot-cli (macOS)
• Set path: export COPILOT_CLI_PATH=/path/to/copilot

Pipeline fails with "Cycle detected"

• Check dependsOn - must form valid DAG (no circular dependencies)

Agents hang during initialization

• Each agent needs its own client - per-agent architecture prevents this

See Troubleshooting Guide for more.

Documentation

• Pipeline System - YAML workflows, DAG execution
• Architecture - Design decisions, patterns
• API Reference - Programmatic usage
• Contributing - Development guide

License

MIT - See LICENSE