walkinglabs/learn-harness-engineering

What is this project?

Learn Harness Engineering is a free, open-source course that teaches you how to build the supporting systems that make AI coding agents (like Claude, GPT, or Codex) work reliably on real software projects. It's not about writing better prompts — it's about designing the environment the AI operates inside.

The core idea: even the most powerful AI model will fail on real engineering tasks if you don't give it a proper structure. The course shows you how to build that structure, which it calls a harness.

What does a harness do?

The README explains that a harness has five subsystems, each with a specific job:

Instructions – Tell the agent what to do, in what order, and what to read before starting. Not one giant file, but a progressive disclosure structure the agent navigates on demand.
State – Track what's been done, what's in progress, and what's next. Saved to disk so the next session picks up exactly where the last one left off.
Verification – Only a passing test suite counts as evidence. The agent cannot claim success without runnable proof.
Scope – Constrain the agent to one feature at a time. No overreach, no half-finishing multiple things, no rewriting the feature list to hide unfinished work.
Session Lifecycle – Initialize at the start, clean up at the end, and leave a clean restart path for the next session.

The README includes a diagram showing how these fit together, and a text version that says: "The MODEL decides what code to write. The HARNESS governs when, where, and how it writes it."

What's in the course?

The course is divided into three parts:

Lectures – 14 conceptual units explaining the theory behind harness engineering.
Projects – 8 hands-on projects where you build an agentic workspace from scratch.
Resource Library – Copy-ready templates (like AGENTS.md, feature_list.json, init.sh) you can drop into your own repositories.

There's also a skills section, including a harness-creator skill that can scaffold a production-grade harness for your project in minutes.

How can you use it right away?

The README offers a quick start: if you're already using a coding agent, you can improve it today by adding four files to your project root:

AGENTS.md – the agent's operating manual
init.sh – runs install + verify + start
feature_list.json – what features exist, which are done
claude-progress.md – a session progress log (the name is historical but it's agent-agnostic; any agent can use it)

These files live inside your repo, so every session starts from the same state. The README says this alone makes agent sessions "significantly more stable than running on prompts alone."

What's the capstone project?

All six (now eight) course projects revolve around the same product: a real app. The README cuts off mid-sentence, but it clearly states the projects are built around a single real application.

What's new recently?

The README highlights two major updates:

August 2026: Added a new section on Frontier Harness Design Breakdowns – four deep dives into how real products (Pi, Claude Code, Codex, DeepSeek) build their harnesses, applying the course's five-subsystem framework. Also added Lecture 14 on Graph Engineering and Project 08 on drawing your workflow as a graph.
July 2026: Added Lecture 13 on Loop Engineering and Project 07 on building your first automated loop, plus code templates for goal loops, timer loops, and maker-checker loops.

Other details

Languages: The course is available in 15 languages (English, Chinese, Japanese, Korean, Spanish, French, Russian, German, Arabic, Vietnamese, Uzbek, Turkish, Portuguese-BR, Ukrainian, and more).
PDFs: There's a build pipeline to generate PDF coursebooks (npm run pdf:build), with output to artifacts/pdfs/.
License: MIT.
Community: There's a Discord link for joining the community.

Why does this course exist?

The README explains the motivation: models can write code, but can they reliably complete real engineering tasks inside real repositories, over multiple sessions, without constant human supervision? Right now, the answer is "not without a harness." The course teaches you how to build that harness, so you can trust your AI agent to do the work while you verify the results instead of cleaning up after it.

In short: Learn Harness Engineering is a practical, project-based course that gives you the framework and tools to turn an AI coding agent from a sometimes-helpful assistant into a dependable teammate.