Learn Harness Engineering (2026): The Free Agent Harness Course
Short answer: Learn Harness Engineering is an open-source course by the GitHub organization walkinglabs about "building the environment, state management, verification, and control mechanisms that make AI coding agents work reliably." It has 14 lectures and 8 hands-on projects, covers Claude Code and Codex, and includes breakdowns of how Pi, Claude Code, Codex and DeepSeek design their harnesses. It is MIT-licensed and available in 15 languages.
Unofficial guide. Antigravity Skills is not affiliated with walkinglabs. The official sources are the GitHub repository and the course site. Facts below were checked on November 13, 2026. If you searched for "learnharness": learnharness.com is a parked domain that is unrelated to this course.
Want the patterns without building them yourself? Our Project Memory Skill gives your agent a clean progress file between sessions, and the Engineering Team Skill adds a review and QA loop. Each download uses 1 credit, and new accounts get 1 free credit.
At a glance
| What it is | A free course on harness engineering for AI coding agents |
| Made by | walkinglabs (GitHub organization) |
| License | MIT |
| Lectures | 14 (L01–L14) |
| Projects | 8 (P01–P08), all on one Electron knowledge-base app |
| Agents covered | Claude Code and Codex, plus breakdowns of Pi, Claude Code, Codex and DeepSeek |
| Languages | 15, including English, Simplified and Traditional Chinese, Japanese, Korean and Spanish |
| GitHub stars | About 24.4k (November 13, 2026) |
What "harness engineering" means
A harness is everything around the model that decides whether an agent finishes the job: the instructions it reads, the state it keeps between sessions, how it checks its own work, and what it is allowed to do. The course's own lecture titles describe the problems it solves, for example "Why Capable Agents Still Fail", "Why One Giant Instruction File Fails" and "Why Every Session Must Leave a Clean State".
What's in the course
- 14 lectures that move from "What a Harness Actually Is" to "From Manual Prompting to Autonomous Loops" and "From Single Loops to Graph Engineering".
- 8 projects that build and harden one personal knowledge-base app, including "Build Your First Automated Loop" and "Draw Your Workflow as a Graph".
- Frontier harness breakdowns of how Pi, Claude Code, Codex and DeepSeek design their harnesses.
- A resource library of templates such as
AGENTS.md, a feature list and a progress file. - A
harness-creatorskill that the README says "can help you scaffold a production-grade harness".
The repository's curriculum section and its badges disagree slightly on the counts (13 units and 7 projects versus 14 lectures and 8 projects). The course site lists 14 lectures.
Where to start
- Read lectures L01 and L02 on the course site to see what a harness is.
- Clone the repository and follow the first project.
- Add the habits to your own repo: one short instruction file, a progress file the agent updates, and a check the agent runs before it says it is done.
FAQ
Is Learn Harness Engineering free?
Yes. The course materials are open source under the MIT license, and the course site does not list any price.
Who made Learn Harness Engineering?
The GitHub organization walkinglabs. The pages don't name an individual author.
Does it work with Claude Code and Codex?
Yes. The course site names Codex and Claude Code, and the repository also mentions OpenHands and Antigravity.
Is it the same as "Learn Claude Code"?
No. "Learn Claude Code" by shareAI-lab is a separate open-source project focused on Claude Code itself. Learn Harness Engineering is a course on making coding agents in general reliable.
What is learnharness.com?
It is a parked domain for sale and is not connected to this course. The course lives on GitHub and GitHub Pages.
Related on Antigravity Skills
- Project Memory Skill: a plain-markdown memory and progress file in your repo
- Engineering Team Skill: a review and QA loop for every feature
- Parallel Agents Skill: run several agents at once with git worktrees
- AGENTS.md Skill: write a clear instruction file for your agents