alinaqi/maggy

What started as an opinionated Claude Code setup kit is now an autonomous AI engineering command center

What it solves

Claude Code is powerful but often uses expensive models for trivial tasks, loses state across sessions, and lacks enforcement for code quality, test coverage, and architectural compliance. Maggy and its Bootstrap config pack provide a structured engineering system that adds TDD enforcement, persistent memory, and cost-aware model routing to the AI coding workflow.

How it works

The project consists of two main components:

  1. Claude Bootstrap: A set of skills, hooks, and rules installed directly into the Claude Code configuration. It enforces quality gates (e.g., function length limits) and TDD (tests must pass before a task is marked done).
  2. Maggy Harness: A local FastAPI server and web dashboard that extends Bootstrap with advanced features. It uses a semantic "blast score" to route requests across 13 different LLM tiers (from local Qwen3 to Claude Opus) to optimize for cost and capability. It also includes "Polyphony" for running multiple agents in isolated Docker containers to prevent file conflicts.

Who it’s for

  • Solo engineers who want automated TDD and persistent memory without changing their existing Claude Code workflow.
  • Teams needing a single dashboard to route work across multiple models (Claude, DeepSeek, Gemini, etc.) while managing costs.
  • Platform engineers looking for a reference implementation of intent-tracking and plugin-based AI developer tooling.

Highlights

  • 13-Tier Model Routing: Automatically selects the cheapest capable model based on task complexity.
  • TDD & Quality Gates: Hard enforcement of test passes and architectural rules (e.g., max 20 lines per function).
  • Polyphony: Docker-isolated parallel agent execution to allow multiple agents to work on the same repo without collisions.
  • Cortex MCP: A code intelligence server providing AST extraction and bidirectional traversal via a SQLite graph store.
  • Telos: An intent-fidelity scoring system that measures if code actually fulfills its original intent beyond just passing tests.
  • Mnemos & Engram: Multi-dimensional memory systems that persist architectural knowledge and task state across sessions.

Related

  • Project
    modu-ai/moai-adk

    A verification-driven orchestration harness for Claude Code that ensures AI-generated code is trustworthy by enforcing evidence-based verification and structured agent orchestration.

  • Dispatch
    Scaling AI Coding: How Claude Code Navigates Large Codebases

    An exploration of agentic search and the 'harness' architecture used by Claude Code to manage multi-million line monorepos and legacy systems without relying on traditional RAG indexing.

  • Project
    Chachamaru127/claude-code-harness

    A disciplined delivery loop and safety layer for AI coding agents that enforces a structured Plan-Work-Review-Release workflow to prevent agent drift.

  • Project
    shareAI-lab/learn-claude-code

    A 17‑lesson tutorial that teaches you how to build the surrounding “harness” code for an LLM‑based coding agent (Claude). It walks through adding tools, permissions, context management, memory, task scheduling, multi‑agent coordination, and goal‑driven termination, all around a simple agent‑loop that lets the model decide when to call tools.

  • Dispatch
    Mastering Claude Code: From Prompting to Programmable Engineering

    A deep dive into transforming Claude Code from a simple chatbot into a sophisticated autonomous agent using custom skills, subagents, and the Model Context Protocol.