A SPEC-first agentic orchestration framework for Claude Code that drives high-quality software delivery through the Harness Engineering paradigm.
MoAI-ADK (MoAI Agentic Development Kit) is a strategic orchestration framework built exclusively for Claude Code, centered on the Harness Engineering paradigm of "humans steer, agents execute." It uses SPEC documents as the single source of truth, automatically switching between TDD (new projects) and DDD (legacy projects) methodologies based on test coverage, and orchestrating 27 specialized agents through the full coding-testing-fixing cycle.
The framework embeds the TRUST 5 quality assurance system (Test/Readability/Uniformity/Security/Traceability) with a Self-Verify Loop for autonomous "code→test→fail→fix→pass" cycles, targeting 85–100% test coverage. 47 Skills use 3-level progressive disclosure to save ~67% token consumption, with dual execution modes (Sub-Agent and Agent Teams) that auto-scale into isolated git worktrees for complex tasks.
Originally written in Python (~73K lines), the project has been fully rewritten in Go (38.7K+ lines, 38 packages), currently at v2.12.0 with 601+ commits. It supports 16 programming languages with automatic toolchain detection, intelligently allocates models (Opus/Sonnet/Haiku) based on Claude Code subscription plans, and provides cross-session memory persistence. Use cases include greenfield development, legacy refactoring, large-scale parallel feature development, batch GitHub Issue resolution, and PR review.
CLI Commands: moai init (initialize), moai update (reconfigure), moai migrate agency (migrate legacy commands).
Claude Code Slash Commands: /moai project (generate project docs), /moai plan (create SPEC), /moai run SPEC-XXX (execute implementation), /moai sync (sync docs and create PR), /moai loop (self-verify loop), /moai codemaps (code architecture maps), /moai clean (garbage collection), /moai mx (large-scale refactoring), /moai simplify (auto quality review), /moai github pr/issues (PR review & Issue fixing), /batch (parallel scaling).
Configuration: Quality config at .moai/config/sections/quality.yaml, local settings via settings.local.json (not committed to Git), model policy configured through moai update.
Unconfirmed: Minimum Go version (README states 1.26+, possibly forward-looking or a typo); whether GitHub integration requires a Token; Python version status.