[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-74805":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":10,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":24,"hasPages":24,"topics":25,"createdAt":10,"pushedAt":10,"updatedAt":26,"readmeContent":27,"aiSummary":28,"trendingCount":16,"starSnapshotCount":16,"syncStatus":29,"lastSyncTime":30,"discoverSource":31},74805,"learn-harness-engineering","walkinglabs\u002Flearn-harness-engineering","walkinglabs","Harness engineering beginner tutorial, from 0 to 1","https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002F",null,"TypeScript",8186,853,33,9,0,354,867,2004,1062,39.79,false,"main",true,[],"2026-06-12 02:03:28","\u003Cp align=\"center\">\n  \u003Ca href=\"README.md\">\u003Cimg alt=\"English\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FEN-English-blue?style=flat-square\">\u003C\u002Fa>\n  \u003Ca href=\"docs-readme\u002Fzh-CN\u002FREADME.md\">\u003Cimg alt=\"简体中文\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FZH-简体中文-red?style=flat-square\">\u003C\u002Fa>\n  \u003Ca href=\"docs-readme\u002Fzh-TW\u002FREADME.md\">\u003Cimg alt=\"繁體中文\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FZH--TW-繁體中文-orange?style=flat-square\">\u003C\u002Fa>\n  \u003Ca href=\"docs-readme\u002Fja-JP\u002FREADME.md\">\u003Cimg alt=\"日本語\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FJA-日本語-green?style=flat-square\">\u003C\u002Fa>\n  \u003Ca href=\"docs-readme\u002Fko-KR\u002FREADME.md\">\u003Cimg alt=\"한국어\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FKO-한국어-blueviolet?style=flat-square\">\u003C\u002Fa>\n  \u003Ca href=\"docs-readme\u002Fes-ES\u002FREADME.md\">\u003Cimg alt=\"Español\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FES-Español-yellow?style=flat-square\">\u003C\u002Fa>\n  \u003Ca href=\"docs-readme\u002Ffr-FR\u002FREADME.md\">\u003Cimg alt=\"Français\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FFR-Français-007EC6?style=flat-square\">\u003C\u002Fa>\n  \u003Ca href=\"docs-readme\u002Fru-RU\u002FREADME.md\">\u003Cimg alt=\"Русский\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FRU-Русский-informational?style=flat-square\">\u003C\u002Fa>\n  \u003Ca href=\"docs-readme\u002Fde-DE\u002FREADME.md\">\u003Cimg alt=\"Deutsch\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDE-Deutsch-2EA043?style=flat-square\">\u003C\u002Fa>\n  \u003Ca href=\"docs-readme\u002Far-SA\u002FREADME.md\">\u003Cimg alt=\"العربية\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FAR-العربية-success?style=flat-square\">\u003C\u002Fa>\n  \u003Ca href=\"docs-readme\u002Fvi-VN\u002FREADME.md\">\u003Cimg alt=\"Tiếng Việt\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FVI-Tiếng_Việt-cc6699?style=flat-square\">\u003C\u002Fa>\n  \u003Ca href=\"docs-readme\u002Fuz-UZ\u002FREADME.md\">\u003Cimg alt=\"Oʻzbekcha\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FUZ-Oʻzbekcha-1A8BBA?style=flat-square\">\u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Ch1 align=\"center\">Learn Harness Engineering\u003C\u002Fh1>\n\n\u003Cp align=\"center\">\u003Cstrong>A project-based course on building the environment, state management, verification, and control mechanisms that make AI coding agents work reliably.\u003C\u002Fstrong>\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLectures-12-blue?style=flat-square\" alt=\"12 Lectures\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FProjects-6-green?style=flat-square\" alt=\"6 Projects\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLanguages-12-yellow?style=flat-square\" alt=\"12 Languages\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-lightgrey?style=flat-square\" alt=\"MIT License\">\n\u003C\u002Fp>\n\n> **Globe icon** This course is available in **12 languages**: English, 简体中文, 繁體中文, 日本語, 한국어, Español, Français, Русский, Deutsch, العربية, Tiếng Việt, Oʻzbekcha. Choose your language from the badges above.\n\nLearn Harness Engineering is a course dedicated to the engineering of AI coding agents. We have deeply studied and synthesized the most advanced Harness Engineering theories and practices in the industry. Our core references include:\n\n- [OpenAI: Harness engineering: leveraging Codex in an agent-first world](https:\u002F\u002Fopenai.com\u002Findex\u002Fharness-engineering\u002F)\n- [Anthropic: Effective harnesses for long-running agents](https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Feffective-harnesses-for-long-running-agents)\n- [Anthropic: Harness design for long-running application development](https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Fharness-design-long-running-apps)\n- [Awesome Harness Engineering](https:\u002F\u002Fgithub.com\u002Fwalkinglabs\u002Fawesome-harness-engineering)\n\n> **Quick start?** The [`skills\u002Fharness-creator\u002F`](.\u002Fskills\u002F) skill can help you scaffold a production-grade harness (AGENTS.md, feature lists, init.sh, verification workflows) for your own project in minutes.\n\n---\n\n## Table of Contents\n\n- [✨ Visual Preview](#-visual-preview)\n- [What Harness Engineering Actually Means](#what-harness-engineering-actually-means)\n- [Quick Start: Improve Your Agent Today](#quick-start-improve-your-agent-today)\n- [Capstone Project: A Real App](#capstone-project-a-real-app)\n- [Learning Path](#learning-path)\n- [Syllabus](#syllabus)\n- [Skills](#skills)\n- [Other Courses](#other-courses)\n\n---\n\n## ✨ Visual Preview\n\n### 🏠 Course Homepage\n> A comprehensive course outline and introduction to core philosophies, providing a clear path to get started.\n\n![Course homepage preview](.\u002Fdocs\u002Fpublic\u002Fscreenshots\u002Freadme\u002Fen-home.png)\n\n### 📖 Immersive Lectures\n> Deep dives into real-world pain points and hands-on projects (like Project 01) for an immersive learning experience.\n\n![Course lecture preview](.\u002Fdocs\u002Fpublic\u002Fscreenshots\u002Freadme\u002Fen-lecture-01.png)\n\n### 🗂️ Ready-to-Use Resource Library\n> Templates and reference configurations designed to solve common pitfalls in multi-turn AI agent development, such as context loss and premature task completion.\n\n![Resource library preview](.\u002Fdocs\u002Fpublic\u002Fscreenshots\u002Freadme\u002Fen-resources.png)\n\n## PDF Coursebooks\n\nThe repository now includes a PDF build pipeline for the course content.\n\n- Run `npm run pdf:build` to generate the currently configured PDF coursebooks locally.\n- Output files are written to `artifacts\u002Fpdfs\u002F`.\n- Run `npm run screenshots:readme` if you want to refresh the README preview images.\n- GitHub Actions workflow [`release-course-pdfs.yml`](.\u002F.github\u002Fworkflows\u002Frelease-course-pdfs.yml) can build the PDFs and publish them to GitHub Releases.\n\n---\n\n## The Model Is Smart, The Harness Makes It Reliable\n\nThere's a hard truth most people learn the hard way: **the strongest model in the world will still fail on real engineering tasks if you don't build a proper environment around it.**\n\nYou've probably seen this yourself. You give Claude or GPT a task in your repo. It starts well — reads files, writes code, looks productive. Then something goes wrong. It skips a step. It breaks a test. It says \"done\" but nothing actually works. You spend more time cleaning up than if you'd done it yourself.\n\nThis isn't a model problem. It's a harness problem.\n\nThe evidence is clear. Anthropic ran a controlled experiment: same model (Opus 4.5), same prompt (\"build a 2D retro game editor\"). Without a harness, it spent $9 in 20 minutes and produced something that didn't work. With a full harness (planner + generator + evaluator), it spent $200 in 6 hours and built a game you could actually play. The model didn't change. The harness did.\n\nOpenAI reported the same thing with Codex: in a well-harnessed repository, the same model goes from \"unreliable\" to \"reliable.\" Not a marginal improvement — a qualitative shift.\n\n**This course teaches you how to build that environment.**\n\n```text\n                    THE HARNESS PATTERN\n                    ====================\n\n    You --> give task --> Agent reads harness files --> Agent executes\n                                                        |\n                                              harness governs every step:\n                                              |\n                                              +--> Instructions: what to do, in what order\n                                              +--> Scope:       one feature at a time, no overreach\n                                              +--> State:       progress log, feature list, git history\n                                              +--> Verification: tests, lint, type-check, smoke runs\n                                              +--> Lifecycle:   init at start, clean state at end\n                                              |\n                                              v\n                                         Agent stops only when\n                                         verification passes\n```\n\n---\n\n## What Harness Engineering Actually Means\n\nHarness engineering is about building a complete working environment around the model so it produces reliable results. It's not about writing better prompts. It's about designing the system the model operates inside.\n\nA harness has five subsystems:\n\n```text\n    ┌─────────────────────────────────────────────────────────────────┐\n    │                        THE HARNESS                              │\n    │                                                                 │\n    │   ┌──────────────┐  ┌──────────────┐  ┌──────────────────────┐ │\n    │   │ Instructions  │  │    State     │  │   Verification       │ │\n    │   │              │  │              │  │                      │ │\n    │   │ AGENTS.md    │  │ progress.md  │  │ tests + lint         │ │\n    │   │ CLAUDE.md    │  │ feature_list │  │ type-check           │ │\n    │   │ feature_list │  │ git log      │  │ smoke runs           │ │\n    │   │ docs\u002F        │  │ session hand │  │ e2e pipeline         │ │\n    │   └──────────────┘  └──────────────┘  └──────────────────────┘ │\n    │                                                                 │\n    │   ┌──────────────┐  ┌──────────────────────────────────────┐   │\n    │   │    Scope     │  │         Session Lifecycle             │   │\n    │   │              │  │                                      │   │\n    │   │ one feature  │  │ init.sh at start                     │   │\n    │   │ at a time   │  │ clean-state checklist at end          │   │\n    │   │ definition   │  │ handoff note for next session        │   │\n    │   │ of done      │  │ commit only when safe to resume      │   │\n    │   └──────────────┘  └──────────────────────────────────────┘   │\n    │                                                                 │\n    └─────────────────────────────────────────────────────────────────┘\n\n    The MODEL decides what code to write.\n    The HARNESS governs when, where, and how it writes it.\n    The harness doesn't make the model smarter.\n    It makes the model's output reliable.\n```\n\nEach subsystem has one job:\n\n- **Instructions** — Tell the agent what to do, in what order, and what to read before starting. Not one giant file; a progressive disclosure structure the agent navigates on demand.\n- **State** — Track what's been done, what's in progress, and what's next. Persisted to disk so the next session picks up exactly where the last one left off.\n- **Verification** — Only a passing test suite counts as evidence. The agent cannot declare victory without runnable proof.\n- **Scope** — Constrain the agent to one feature at a time. No overreach. No half-finishing three things. No rewriting the feature list to hide unfinished work.\n- **Session Lifecycle** — Initialize at the start. Clean up at the end. Leave a clean restart path for the next session.\n\n---\n\n## Why This Course Exists\n\nThe question isn't \"can models write code?\" They can. The question is: **can they reliably complete real engineering tasks inside real repositories, over multiple sessions, without constant human supervision?**\n\nRight now, the answer is: not without a harness.\n\n```text\n    WITHOUT HARNESS                          WITH HARNESS\n    ==============                          ============\n\n    Session 1: agent writes code            Session 1: agent reads instructions\n              agent breaks tests                      agent runs init.sh\n              agent says \"done\"                       agent works on one feature\n              you fix it manually                     agent verifies before claiming done\n                                                       agent updates progress log\n    Session 2: agent starts fresh                    agent commits clean state\n              agent has no memory\n              of what happened before         Session 2: agent reads progress log\n              agent re-does work                       agent picks up exactly where it left off\n              or does something else entirely          agent continues the unfinished feature\n              you fix it again                         you review, not rescue\n\n    Result: you spend more time                  Result: agent does the work,\n            cleaning up than if you                      you verify the result\n            did it yourself\n```\n\nThe questions this course actually cares about:\n\n- Which harness designs improve task completion rates?\n- Which designs reduce rework and incorrect completions?\n- Which mechanisms keep long-running tasks progressing steadily?\n- Which structures keep the system maintainable after multiple agent runs?\n\n---\n\n## Course Curriculum & Documentation\n\nFor the full course materials, please visit the **[Documentation Website](https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002F)**.\n\nThe curriculum is divided into three parts:\n\n1. **Lectures**: 12 conceptual units explaining the theory behind harness engineering.\n2. **Projects**: 6 hands-on projects where you build an agentic workspace from scratch.\n3. **Resource Library**: Copy-ready templates (`AGENTS.md`, `feature_list.json`, `init.sh`, etc.) to use in your own repositories today.\n\n---\n\n## Quick Start: Improve Your Agent Today\n\nYou don't need to read all 12 lectures before you start getting value. If you're already using a coding agent on a real project, here's how to improve it right now.\n\nThe idea is simple: instead of just writing prompts, give your agent a set of structured files that define what to do, what's been done, and how to verify the work. These files live inside your repo, so every session starts from the same state.\n\n```text\n    YOUR PROJECT ROOT\n    ├── AGENTS.md              \u003C-- the agent's operating manual\n    ├── CLAUDE.md              \u003C-- (alternative, if using Claude Code)\n    ├── init.sh                \u003C-- runs install + verify + start\n    ├── feature_list.json      \u003C-- what features exist, which are done\n    ├── claude-progress.md     \u003C-- what happened each session\n    └── src\u002F                   \u003C-- your actual code\n```\n\nGrab the starter templates from the [Resource Library](https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002Fen\u002Fresources\u002F) and drop them into your project. That's it. Four files, and your agent sessions will already be significantly more stable than running on prompts alone.\n\n---\n\n## Capstone Project: A Real App\n\nAll six course projects revolve around the same product: **an Electron-based personal knowledge base desktop app**.\n\n```text\n    ┌─────────────────────────────────────────────────────┐\n    │               Knowledge Base Desktop App            │\n    │                                                     │\n    │  ┌──────────────┐  ┌──────────────────────────────┐│\n    │  │ Document List │  │       Q&A Panel              ││\n    │  │              │  │                              ││\n    │  │ doc-001.md   │  │  Q: What is harness eng?    ││\n    │  │ doc-002.md   │  │  A: The environment built    ││\n    │  │ doc-003.md   │  │     around an agent model... ││\n    │  │ ...          │  │     [citation: doc-002.md]   ││\n    │  └──────────────┘  └──────────────────────────────┘│\n    │                                                     │\n    │  ┌─────────────────────────────────────────────────┐│\n    │  │ Status Bar: 42 docs | 38 indexed | last sync 3m ││\n    │  └─────────────────────────────────────────────────┘│\n    └─────────────────────────────────────────────────────┘\n\n    Core features:\n    ├── Import local documents\n    ├── Manage a document library\n    ├── Process and index documents\n    ├── Run AI-powered Q&A over imported content\n    └── Return grounded answers with citations\n```\n\nThis project was chosen because it combines strong practical value, enough real-world product complexity, and a good setting for observing before\u002Fafter harness improvements.\n\nEach course project's starter\u002Fsolution is a complete copy of this Electron app at that evolutionary stage. P(N+1)'s starter is derived from P(N)'s solution — the app evolves as your harness skills grow.\n\n---\n\n## Learning Path\n\nThe course is designed to be done in order. Each phase builds on the last.\n\n```text\n    Phase 1: SEE THE PROBLEM              Phase 2: STRUCTURE THE REPO\n    ========================              ==========================\n\n    L01  Strong models ≠ reliable         L03  Repository as single\n         execution                              source of truth\n    L02  What harness actually means\n                                       L04  Split instructions across\n         |                                   files, not one giant file\n         v\n    P01  Prompt-only vs.                       |\n         rules-first comparison                v\n                                               P02  Agent-readable workspace\n\n\n    Phase 3: CONNECT SESSIONS             Phase 4: FEEDBACK & SCOPE\n    ==========================           =========================\n\n    L05  Keep context alive               L07  Draw clear task boundaries\n         across sessions\n                                       L08  Feature lists as harness\n    L06  Initialize before every               primitives\n         agent session\n                                               |\n         |                                     v\n         v                                     P04  Runtime feedback to\n    P03  Multi-session continuity                   correct agent behavior\n\n\n    Phase 5: VERIFICATION                 Phase 6: PUT IT ALL TOGETHER\n    =====================                 ============================\n\n    L09  Stop agents from                 L11  Make agent's runtime\n         declaring victory early               observable\n\n    L10  Full-pipeline run =              L12  Clean handoff at end of\n         real verification                      every session\n\n         |                                     |\n         v                                     v\n    P05  Agent verifies its own work       P06  Build a complete harness\n                                               (capstone project)\n```\n\nEach phase takes about a week if you're going part-time. If you want to go faster, phases 1–3 can be done in a long weekend.\n\n---\n\n## Syllabus\n\n### Lectures — 12 conceptual units, each answering one core question\n\n*Read the full text for each lecture on the [Documentation Website](https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002F).*\n\n| Session | Question | Core Idea |\n|---------|----------|-----------|\n| [L01](.\u002Fdocs\u002Flectures\u002Flecture-01-why-capable-agents-still-fail\u002Findex.md) | Why do strong models still fail on real tasks? | The capability gap between benchmarks and real engineering |\n| [L02](.\u002Fdocs\u002Flectures\u002Flecture-02-what-a-harness-actually-is\u002Findex.md) | What does \"harness\" actually mean? | Five subsystems: instructions, state, verification, scope, lifecycle |\n| [L03](.\u002Fdocs\u002Flectures\u002Flecture-03-why-the-repository-must-become-the-system-of-record\u002Findex.md) | Why must the repo be the single source of truth? | If the agent can't see it, it doesn't exist |\n| [L04](.\u002Fdocs\u002Flectures\u002Flecture-04-why-one-giant-instruction-file-fails\u002Findex.md) | Why does one giant instruction file fail? | Progressive disclosure: give a map, not an encyclopedia |\n| [L05](.\u002Fdocs\u002Flectures\u002Flecture-05-why-long-running-tasks-lose-continuity\u002Findex.md) | Why do long-running tasks lose continuity? | Persist progress to disk; pick up where you left off |\n| [L06](.\u002Fdocs\u002Flectures\u002Flecture-06-why-initialization-needs-its-own-phase\u002Findex.md) | Why does initialization need its own phase? | Verify the environment is healthy before the agent starts work |\n| [L07](.\u002Fdocs\u002Flectures\u002Flecture-07-why-agents-overreach-and-under-finish\u002Findex.md) | Why do agents overreach and under-finish? | One feature at a time; explicit definition of done |\n| [L08](.\u002Fdocs\u002Flectures\u002Flecture-08-why-feature-lists-are-harness-primitives\u002Findex.md) | Why are feature lists harness primitives? | Machine-readable scope boundaries the agent can't ignore |\n| [L09](.\u002Fdocs\u002Flectures\u002Flecture-09-why-agents-declare-victory-too-early\u002Findex.md) | Why do agents declare victory too early? | Verification gaps: confidence ≠ correctness |\n| [L10](.\u002Fdocs\u002Flectures\u002Flecture-10-why-end-to-end-testing-changes-results\u002Findex.md) | Why does end-to-end testing change results? | Only a full-pipeline run counts as real verification |\n| [L11](.\u002Fdocs\u002Flectures\u002Flecture-11-why-observability-belongs-inside-the-harness\u002Findex.md) | Why does observability belong inside the harness? | If you can't see what the agent did, you can't fix what it broke |\n| [L12](.\u002Fdocs\u002Flectures\u002Flecture-12-why-every-session-must-leave-a-clean-state\u002Findex.md) | Why must every session leave a clean state? | The next session's success depends on this session's cleanup |\n\n### Projects — 6 hands-on projects applying lecture methods to the same Electron app\n\n| Project | What You Do | Harness Mechanism |\n|---------|------------|-------------------|\n| [P01](.\u002Fdocs\u002Fprojects\u002Fproject-01-baseline-vs-minimal-harness\u002Findex.md) | Run the same task twice: prompt-only vs. rules-first | Minimal harness: AGENTS.md + init.sh + feature_list.json |\n| [P02](.\u002Fdocs\u002Fprojects\u002Fproject-02-agent-readable-workspace\u002Findex.md) | Restructure the repo so the agent can read it | Agent-readable workspace + persistent state files |\n| [P03](.\u002Fdocs\u002Fprojects\u002Fproject-03-multi-session-continuity\u002Findex.md) | Make the agent pick up from where it left off | Progress log + session handoff + multi-session continuity |\n| [P04](.\u002Fdocs\u002Fprojects\u002Fproject-04-incremental-indexing\u002Findex.md) | Stop the agent from doing too much or too little | Runtime feedback + scope control + incremental indexing |\n| [P05](.\u002Fdocs\u002Fprojects\u002Fproject-05-grounded-qa-verification\u002Findex.md) | Make the agent verify its own work | Self-verification + grounded Q&A + evidence-based completion |\n| [P06](.\u002Fdocs\u002Fprojects\u002Fproject-06-runtime-observability-and-debugging\u002Findex.md) | Build a complete harness from scratch (capstone) | Full harness: all mechanisms + observability + ablation study |\n\n```text\n    PROJECT EVOLUTION\n    =================\n\n    P01  Prompt-only vs. rules-first       You see the problem\n     |\n     v\n    P02  Agent-readable workspace           You restructure the repo\n     |\n     v\n    P03  Multi-session continuity           You connect sessions\n     |\n     v\n    P04  Runtime feedback & scope           You add feedback loops\n     |\n     v\n    P05  Self-verification                  You make the agent check itself\n     |\n     v\n    P06  Complete harness (capstone)        You build the full system\n\n    Each project's solution becomes the next project's starter.\n    The app evolves. Your harness skills grow with it.\n```\n\n### Resource Library\n\n- [English](https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002Fen\u002Fresources\u002F) — templates, checklists, and method references\n- [简体中文](https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002Fzh\u002Fresources\u002F) — 中文模板、清单和方法参考\n- [繁體中文](https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002Fzh-TW\u002Fresources\u002F) — 繁體中文範本、清單和方法參考\n- [日本語](https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002Fja\u002Fresources\u002F) — テンプレート、チェックリスト、方法リファレンス\n- [한국어](https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002Fko\u002Fresources\u002F) — 템플릿, 체크리스트, 방법 참고 자료\n- [Español](https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002Fes\u002Fresources\u002F) — plantillas, listas de verificación y referencias\n- [Français](https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002Ffr\u002Fresources\u002F) — modèles, listes de contrôle et références\n- [Русский](https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002Fru\u002Fresources\u002F) — шаблоны, чек-листы и справочники\n- [Deutsch](https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002Fde\u002Fresources\u002F) — Vorlagen, Checklisten und Referenzen\n- [العربية](https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002Far\u002Fresources\u002F) — قوالب، قوائم تحقق ومراجع\n- [Tiếng Việt](https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002Fvi\u002Fresources\u002F) — mẫu, danh sách kiểm tra và tài liệu tham khảo\n- [Oʻzbekcha](https:\u002F\u002Fwalkinglabs.github.io\u002Flearn-harness-engineering\u002Fuz\u002Fresources\u002F) — andozalar, tekshiruv roʻyxatlari va maʼlumotnomalar\n\n---\n\n## The Agent Session Lifecycle\n\nOne of the core ideas in this course: **the agent's session should follow a structured lifecycle, not a free-for-all.** Here's what that looks like:\n\n```text\n    AGENT SESSION LIFECYCLE\n    ======================\n\n    ┌──────────────────────────────────────────────────────────────────┐\n    │  START                                                          │\n    │                                                                  │\n    │  1. Agent reads AGENTS.md \u002F CLAUDE.md                           │\n    │  2. Agent runs init.sh (install, verify, health check)          │\n    │  3. Agent reads claude-progress.md (what happened last time)    │\n    │  4. Agent reads feature_list.json (what's done, what's next)    │\n    │  5. Agent checks git log (recent changes)                       │\n    │                                                                  │\n    │  SELECT                                                          │\n    │                                                                  │\n    │  6. Agent picks exactly ONE unfinished feature                   │\n    │  7. Agent works only on that feature                             │\n    │                                                                  │\n    │  EXECUTE                                                         │\n    │                                                                  │\n    │  8. Agent implements the feature                                 │\n    │  9. Agent runs verification (tests, lint, type-check)           │\n    │  10. If verification fails: fix and re-run                      │\n    │  11. If verification passes: record evidence                    │\n    │                                                                  │\n    │  WRAP UP                                                         │\n    │                                                                  │\n    │  12. Agent updates claude-progress.md                           │\n    │  13. Agent updates feature_list.json                            │\n    │  14. Agent records what's still broken or unverified            │\n    │  15. Agent commits (only when safe to resume)                   │\n    │  16. Agent leaves clean restart path for next session           │\n    │                                                                  │\n    └──────────────────────────────────────────────────────────────────┘\n\n    The harness governs every transition in this lifecycle.\n    The model decides what code to write at each step.\n    Without the harness, step 9 becomes \"agent says it looks fine.\"\n    With the harness, step 9 is \"tests pass, lint is clean, types check.\"\n```\n\n---\n\n## Who This Is For\n\nThis course is for:\n\n- Engineers already using coding agents who want better stability and quality\n- Researchers or builders who want a systematic understanding of harness design\n- Tech leads who need to understand how environment design affects agent performance\n\nThis course is not for:\n\n- People looking for a zero-code AI introduction\n- People who only care about prompts and don't plan to build real implementations\n- Learners not prepared to let agents work inside real repositories\n\n---\n\n## Requirements\n\nThis is a course where you actually run coding agents.\n\nYou need at least one of these tools:\n\n- Claude Code\n- Codex\n- Another IDE or CLI coding agent that supports file editing, command execution, and multi-step tasks\n\nThe course assumes you can:\n\n- Open a local repository\n- Allow the agent to edit files\n- Allow the agent to run commands\n- Inspect output and re-run tasks\n\nIf you don't have such a tool, you can still read the course content, but you won't be able to complete the projects as intended.\n\n---\n\n## Local Preview\n\nThis repository uses VitePress as a documentation viewer.\n\n```sh\nnpm install\nnpm run docs:dev        # Dev server with hot reload\nnpm run docs:build      # Production build\nnpm run docs:preview    # Preview built site\n```\n\nThen open the local URL that VitePress outputs in your browser.\n\n---\n\n## Prerequisites\n\nRequired:\n\n- Familiarity with the terminal, git, and local development environments\n- Ability to read and write code in at least one common application stack\n- Basic software debugging experience (reading logs, tests, and runtime behavior)\n- Enough time to commit to implementation-focused coursework\n\nHelpful but not required:\n\n- Experience with Electron, desktop apps, or local-first tools\n- Background in testing, logging, or software architecture\n- Prior exposure to Codex, Claude Code, or similar coding agents\n\n---\n\n## Core References\n\nPrimary:\n\n- [OpenAI: Harness engineering: leveraging Codex in an agent-first world](https:\u002F\u002Fopenai.com\u002Findex\u002Fharness-engineering\u002F)\n- [Anthropic: Effective harnesses for long-running agents](https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Feffective-harnesses-for-long-running-agents)\n- [Anthropic: Harness design for long-running application development](https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Fharness-design-long-running-apps)\n- [OpenAI: Unrolling the Codex agent loop](https:\u002F\u002Fopenai.com\u002Findex\u002Funrolling-the-codex-agent-loop\u002F)\n- [Anthropic: Demystifying evals for AI agents](https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Fdemystifying-evals-for-ai-agents)\n- [LangChain: Improving Deep Agents with harness engineering](https:\u002F\u002Fwww.langchain.com\u002Fblog\u002Fimproving-deep-agents-with-harness-engineering)\n- [Thoughtworks \u002F Martin Fowler: Harness engineering for coding agent users](https:\u002F\u002Fmartinfowler.com\u002Farticles\u002Fharness-engineering.html)\n- [Cursor: Continually improving our agent harness](https:\u002F\u002Fcursor.com\u002Fblog\u002Fcontinually-improving-agent-harness)\n\nSee the full layered reference list in [`docs\u002Fen\u002Fresources\u002Freference\u002F`](.\u002Fdocs\u002Fen\u002Fresources\u002Freference\u002Findex.md).\n\n---\n\n## Repository Structure\n\n```text\nlearn-harness-engineering\u002F\n├── docs\u002F                          # VitePress documentation site\n│   ├── lectures\u002F                  # 12 lectures (index.md + code\u002F examples)\n│   │   ├── lecture-01-*\u002F\n│   │   ├── lecture-02-*\u002F\n│   │   └── ... (12 total)\n│   ├── projects\u002F                  # 6 project descriptions\n│   │   ├── project-01-*\u002F\n│   │   └── ... (6 total)\n│   └── resources\u002F                 # Multilingual templates & references\n│       ├── en\u002F                    # English templates, checklists, guides\n│       ├── zh\u002F                    # Chinese templates, checklists, guides\n│       ├── ru\u002F                    # Russian templates, checklists, guides\n│       └── vi\u002F                    # Vietnamese templates, checklists, guides\n├── projects\u002F\n│   ├── shared\u002F                    # Shared Electron + TypeScript + React foundation\n│   └── project-NN\u002F               # Per-project starter\u002F and solution\u002F directories\n├── skills\u002F                        # Reusable AI agent skills\n│   └── harness-creator\u002F           # Harness engineering skill\n├── package.json                   # VitePress + dev tooling\n└── CLAUDE.md                      # Claude Code instructions for this repo\n```\n\n---\n\n## How the Course Is Organized\n\n- Each lecture focuses on one question\n- The course includes 6 projects\n- Every project requires the agent to do real work\n- Every project compares weak vs. strong harness results\n- What matters is the measured difference, not how many docs were written\n\n---\n\n## Skills\n\nThis repository also includes reusable AI agent skills that you can install directly into your IDE or agent workspace.\n\n- [**harness-creator**](.\u002Fskills\u002Fharness-creator\u002F): A skill that helps you scaffold a production-grade harness for your own project in minutes.\n\n---\n\n## Other Courses\n\nOur team has also created other courses! Check them out:\n\n[![Hands-on Modern RL](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FHANDS--ON_MODERN_RL-0052cc?style=for-the-badge)](https:\u002F\u002Fgithub.com\u002Fwalkinglabs\u002Fhands-on-modern-rl)\n\n**Hands-on Modern RL**: An open-source, hands-on curriculum bridging the gap from basic RL concepts to LLM alignment, RLVR, and advanced Agentic systems.\n\n---\n\n## Star History\n\n[![Star History Chart](https:\u002F\u002Fapi.star-history.com\u002Fsvg?repos=walkinglabs\u002Flearn-harness-engineering&type=date&legend=top-left)](https:\u002F\u002Fwww.star-history.com\u002F#walkinglabs\u002Flearn-harness-engineering&type=date&legend=top-left)\n\n---\n\n## Acknowledgments\n\nThis course was inspired by and draws ideas from [learn-claude-code](https:\u002F\u002Fgithub.com\u002FshareAI-lab\u002Flearn-claude-code) — a progressive guide to building an agent from scratch, from a single loop to isolated autonomous execution.\n","Learn Harness Engineering 是一个专注于AI编码代理工程的项目基础课程。它涵盖了构建环境、状态管理、验证和控制机制等关键内容，旨在使AI编码代理能够可靠运行。该教程使用TypeScript编写，提供了12个讲座和6个项目实践，支持包括简体中文在内的12种语言版本。适合希望从零开始学习如何设计和实现高效AI代理系统的开发者或工程师，特别是在需要确保AI系统稳定性和可靠性的情境下应用。",2,"2026-06-11 03:50:54","high_star"]