Duo — AI skill for Claude Code
Human-approved engineering decisions for AI coding agents.
How to install Duo
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open kanghyunsoon/duo and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Duo does
Human-approved engineering decisions for AI coding agents. Local-first MCP tools for Codex and Claude Code with evidence-based Git diff reviews.
Alternatives in AI
- Flow Next — Plan-first AI workflow plugin for Claude Code, OpenAI Codex, and Factory Droid 569 ★
- Claude Code Sourcemap Learning Notebook — Reverse-engineering Claude Code's 512K+ lines of TypeScript — architecture, design decisions, and 11 transfera 160 ★
- Linggen Memory — Local-first semantic memory for AI assistants — one store your preferences, decisions and gotchas follow you i 108 ★
README
DUO
[](https://www.npmjs.com/package/@duo-director/cli) [](https://github.com/kanghyunsoon/duo/actions/workflows/ci.yml) [](LICENSE)
**Source repository: [github.com/kanghyunsoon/duo](https://github.com/kanghyunsoon/duo)** (development, issues and releases happen there).
**Decision compliance for coding agents.**
Keep coding agents aligned with human-confirmed engineering decisions. You confirm the decisions. Coding agents such as Codex and Claude Code implement. DUO checks repository changes against those decisions on its own and reports what it found, with file and line evidence. DUO does not write code.
[한국어](README.ko.md) · [Latest release](https://github.com/kanghyunsoon/duo/releases/latest) · [npm](https://www.npmjs.com/package/@duo-director/cli)
New here? [Quick start](#quick-start) · [Verified 0.3.3 walkthrough](docs/examples/decision-review-0.3.3.md) · [Your first contribution](docs/community/START_HERE.md) · [Discussions](https://github.com/kanghyunsoon/duo/discussions)
Why not just AGENTS.md, CLAUDE.md or repository rules?
Instruction files and rules tell an agent what it should do. They are a good place for conventions, and DUO adds a short block to them when you connect an agent.
DUO works on the other side of the change. It keeps human-confirmed Requirements and Decisions as Project Truth in your repository and reviews what actually changed against them, whichever agent made the change. What DUO adds on top of instructions:
- Decision lifecycle. Agents can only propose a Decision. A person confirms or rejects it, and a confirmed Decision is locked with a digest so later edits are visible.
- Deterministic evidence. Review claims point to files, line ranges, Git diff hunks and the Decision itself. The core review does not use a
Related Skills
Context Engineering Protocol
Portable AI coding-agent skills for source-attributed, human-approved context packages.
Hames
A workspace-first harness for approved, evidence-backed AI agent work. Supports Codex and Claude Code.
Simplicity Review
Enforces Law 4 (Verify Before Reporting) of the 7 Laws of AI Agent Discipline. Reviews the current diff for ov
Sloppy
Static analysis for the debt AI agents leave in PHP: 25 rules, Claude Code hooks, git-diff review, a Rector an
Pbi Diff
Generate a human-readable summary of all uncommitted changes in the semantic model since the last git commit.
Unterm
The terminal AI agents can drive — and the cockpit to run them from. 99 MCP methods / 21 namespaces, agent Inb
Related Agents
Kimi Worker
Forward a coding, diagnosis, or review task to Kimi K3 (Moonshot) through the local kimi-worker.sh runtime. Us
Agent Of Empires
Claude Code, OpenCode, Mistral Vibe, Codex CLI, Gemini CLI Coding Agent Terminal Session manager via tmux and
Part 13: Memory Bridge (Give Coding Agents Your Brain)
When you spawn Codex or Claude Code to build something, they start blind. They don't know your architecture de