maxence2997

Mx Harness — Testing skill for Claude Code

Testing community

An engineering harness for AI coding agents — spec-first, TDD, multi-perspective review, clean commits.

How to install Mx Harness

This entry records only its repository, not the path inside it, so there is no exact command to give. Open maxence2997/mx-harness and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Mx Harness does

An engineering harness for AI coding agents — spec-first, TDD, multi-perspective review, clean commits.

Alternatives in Testing

OpenAgentsControl — AI agent framework for plan-first development workflows with approval-based execution 4.8k ★
AI Contribution Skills (Engram) — branch-pr: clean branch and PR workflow 2.5k ★
Spec Driven Develop — Spec-driven development workflow for AI coding agents: architecture-first planning, task decomposition, GitHub 975 ★

README

mx-harness

Your AI agent writes fast. mx-harness makes it engineer properly.

![License](https://img.shields.io/badge/license-MIT-blue) ![Works with](https://img.shields.io/badge/works%20with-Claude%20%7C%20Codex%20%7C%20Copilot%20%7C%20Cursor-blueviolet)

What it looks like

Once installed, you tell the agent your idea — rough or detailed, it asks what it needs.

/mx-flow add Redis caching to the search endpoint

You make a few decisions. The agent handles the rest. [Full walkthrough →](mx-flow/)

Principles

Without a harness, AI agents skip planning, skip tests, and produce unmaintainable diffs. mx-harness wraps the lifecycle into skills the agent must follow.

Principle	Addresses
Spec before code	Vague requirements, hidden assumptions, scope creep
Test-first	Tests written after the fact, missed edge cases
Multi-perspective review	Single-reviewer blind spots, missed SRE concerns
Structured commits	"fix stuff" messages, mixed concerns per commit
Surgical changes	Drive-by edits, inflated diffs, unrelated refactors
Don't assume	Silent guessing on ambiguous specs

The difference

**Without mx-harness**

User:  "Add caching to the search endpoint"
Agent: [writes 200 lines of code]
       [commit: "add cache"]
       [no tests · no design doc · breaks 2 existing behaviours]

**With mx-harness**

User:  /mx-flow "Add caching to the search endpoint"
Agent: → Asks: Redis or in-memory? TTL strategy? Cache invalidation scope?
       → Writes design spec + ADR to ~/.mx/project/search-cache/
       → Waits for approval before touching any code

       → Task 1: Cache interface (testable abstraction)
       → Task 2: Redis adapter
       → Task