harness-engineering

A Claude Code skill for setting up and maintaining agent-first coding environments — the scaffolding, feedback loops, docs, and constraints that let AI coding agents work reliably and autonomously.

What it does

Use this skill when:

• Bootstrapping a new agent-first repo
• Improving CI/feedback loops for coding agents
• Writing or auditing AGENTS.md / ARCHITECTURE.md
• Designing doc structure for agent consumption
• Setting up agent-to-agent review or auto-merge workflows
• Diagnosing why an agent is producing poor results
• Planning recurring cleanup/gardening agent tasks

Install

claude skill install https://github.com/Irreel/harness-engineering

Or clone into your Claude commands directory:

git clone https://github.com/Irreel/harness-engineering ~/.claude/commands/harness-engineering

Usage

Once installed, Claude Code will automatically invoke this skill when relevant. You can also trigger it explicitly:

Use the harness-engineering skill to bootstrap my repo

Contents

• SKILL.md — Main skill definition with workflow, checklist, and key principles
• references/pillars.md — Detailed reference for each of the 7 pillars of a good harness

The 7 Pillars

1. Repository as System of Record — everything the agent needs lives in the repo
2. Tight Feedback Loops — agents see consequences of their actions immediately
3. Enforced Architecture — boundaries enforced centrally, autonomy locally
4. Continuous Garbage Collection — encode cleanup instead of doing it manually
5. Plans as First-Class Artifacts — execution plans versioned in the repo
6. Agent-to-Agent Review — at scale, human review becomes the bottleneck
7. Throughput-First Merge Philosophy — corrections are cheap, waiting is expensive