AgentCraftworks Community Edition

The open protocol layer for agentic DevOps

Quick Start · How It Works · Enterprise · Docs · Contribute

---

The Problem

AI coding agents are powerful — but completely ungoverned. They merge PRs without approval, push to production without validation, and operate at full autonomy with no safety net.

You need a governance layer that grows with your team's trust.

What AgentCraftworks Community Edition Does

AgentCraftworks Community Edition is a GitHub App + MCP server that intercepts every agent action and routes it through configurable Agent Engagement Levels before it reaches your codebase.

SDLC Lifecycle Strategy

AgentCraftworks Community Edition supports teams across multiple SDLC phases, from brand-new product idea to production-ready governance.

• Prototype quickly with low policy friction
• Introduce staging and validation checks as solutions mature
• Enforce productized promotion flow (feature/* -> staging -> main)
• Operate production repos with explicit governance and incident-aware workflows

See docs/SDLC_LIFECYCLE_STRATEGY.md for the lifecycle model and phased policy/infrastructure guidance.

Pull Request / Push Event
         ↓
  AgentCraftworks Community Edition
         ↓
   Agent Engagement Levels (1–5)
    ├── Observer (T1):        Read, view, list
    ├── Advisor (T2):         Comment, suggest
    ├── Peer Programmer (T3): Label, assign, approve, edit file
    ├── Agent Team (T4):      Merge, close, create branch, push commit
    └── Full Agent Team (T5): Deploy, modify CI, orchestrate agents
         ↓
  CODEOWNERS Routing → Assigned Agent
         ↓
   MCP Tool Execution
         ↓
    GitHub Actions

Key Features

Feature	Description
Agent Engagement Levels	5-level governance control (Observer → Full Agent Team) — set per-repo, per-team, per-event type
MCP 6-Tool Interface	Standard MCP server: analyze, fix, review, comment, rollback, escalate
Finite State Machine	Every agent action is a state transition — auditable, reproducible
CODEOWNERS Routing	Events routed to the right agent based on ownership rules
Webhook Handling	Handles GitHub PR, push, issue, and workflow events
GitHub App Scaffold	Drop-in GitHub App: one install, works across all repos in your org

Architecture

graph TD
    GH[GitHub Events] --> WH[Webhook Handler]
    WH --> AD[Engagement Level Router]
    AD -->|Observer / Advisor| OBS[Read & Comment]
    AD -->|Peer Programmer| PR[Label, Assign, Edit]
    AD -->|Agent Team / Full| AUTO[Merge, Deploy, Orchestrate]
    OBS --> MCP[MCP Server]
    PR --> MCP
    AUTO --> MCP
    MCP --> GHA[GitHub Actions]
    MCP --> API[GitHub API]
    GHA --> PROD[Production]

Loading

Quick Start

# Requirements: Node.js 22+, GitHub App credentials
git clone https://github.com/AgentCraftworks/AgentCraftworks-CE.git
cd AgentCraftworks-CE/typescript
npm install

# Configure environment
cp .env.example .env
# Add your GitHub App credentials (see docs/setup.md)

# Build and start
npm run build
npm start

Webhook endpoint: POST /api/webhook
Health check: GET /health
MCP tools: GET /mcp/tools

How It Works

1. Agent Engagement Levels

Every repo gets an engagement level (1–5). The level determines what the agent is permitted to do:

Level	Name	Action Tier	Permitted Actions
1	Observer	T1	Read, view, list
2	Advisor	T2	Comment, suggest
3	Peer Programmer	T3	Label, assign, approve, edit file
4	Agent Team	T4	Merge, close, create branch, push commit
5	Full Agent Team	T5	Deploy, modify CI, orchestrate agents

Environment caps: local=5, dev=5, staging=4, production=3

2. Finite State Machine

Every incoming event follows a deterministic state machine: RECEIVED → CLASSIFIED → GOVERNANCE_CHECK → ROUTED → EXECUTING → COMPLETE

This makes every agent action auditable and reproducible — essential for enterprise compliance.

3. MCP-Compatible

AgentCraftworks Community Edition ships a fully compliant Model Context Protocol (MCP) server. Any MCP-capable AI client (GitHub Copilot, Claude, GPT-4) can connect and use the 6 core tools directly.

Enterprise Edition

AgentCraftworks also powers enterprise deployments with additional incident response automation, self-healing orchestration, and governance monitoring. Learn more at AgentCraftworks.com.

Documentation

• Setup Guide
• Agent Engagement Levels Reference
• MCP Tool Reference
• Architecture Overview
• SDLC Lifecycle Strategy
• Accessibility Capability
• Contributing Guide

Contributing

We welcome contributions! Please read our Contributing Guide first — all contributors must sign our CLA.

# Run tests
cd typescript && npm test

# Lint
npm run lint

License

MIT License — Copyright (c) 2025 AICraftworks LLC

See LICENSE for full text.

---

Built with ❤️ for the agentic DevOps era · Powered by Azure + GitHub Copilot

Enterprise · Issues · Discussions