Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion AGENTS.md
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,7 @@
- [x] Phase 2:形成自己的观点(thinking/,9 篇,持续中)
- [x] Phase 3:选一个小项目实践(practice/,1 个 Ralph Demo)
- [x] Phase 4:记录反馈迭代(feedback/,1 篇,持续中)
- [x] Phase 5:输出可展示的作品(works/,22 篇翻译 + 1 篇原创 + 2 篇外部中文收录)
- [x] Phase 5:输出可展示的作品(works/,24 篇翻译 + 1 篇原创 + 2 篇外部中文收录)

> 进度详情以人类向 README.md 的"学习路线"段为准;本节是给智能体的快照。

Expand Down
24 changes: 13 additions & 11 deletions README.en.md
Original file line number Diff line number Diff line change
@@ -1,8 +1,8 @@
[中文](README.md) | English

![License: MIT](https://img.shields.io/badge/license-MIT-blue)
![Articles](https://img.shields.io/badge/articles-30-green)
![Translations](https://img.shields.io/badge/translations-22-orange)
![Articles](https://img.shields.io/badge/articles-42-green)
![Translations](https://img.shields.io/badge/translations-24-orange)

# Harness Engineering Study Guide

Expand Down Expand Up @@ -112,10 +112,10 @@ harness-engineering/
├── thinking/ # Phase 2: Independent analysis (9 articles)
├── practice/ # Phase 3: Hands-on experiments (1 Ralph Demo)
├── feedback/ # Phase 4: Lessons learned (1 article)
├── works/ # Phase 5: Shareable outputs (22 translations + 1 original + 2 external Chinese captures)
├── works/ # Phase 5: Shareable outputs (24 translations + 1 original + 2 external Chinese captures)
├── tools/ # Tools that reduce the 6 complexity dimensions
├── prompts/ # Validated prompts collection
└── references/ # External resource index (30 articles with deep summaries)
└── references/ # External resource index (42 articles with deep summaries)
```

Each subdirectory has its own `AGENTS.md` explaining its purpose and conventions — a direct practice of the "progressive disclosure" principle from the original article.
Expand All @@ -126,29 +126,31 @@ Each subdirectory has its own `AGENTS.md` explaining its purpose and conventions
- [x] **Phase 2: Form your own opinions** — 9 independent analyses (ongoing)
- [x] **Phase 3: Pick a small project to practice** — Ralph Demo completed (321s, $0.31)
- [x] **Phase 4: Record feedback & iterations** — 1 article (ongoing)
- [x] **Phase 5: Produce shareable work** — 22 professional translations + 1 original synthesis + 2 external Chinese captures
- [x] **Phase 5: Produce shareable work** — 24 professional translations + 1 original synthesis + 2 external Chinese captures

## 📚 Research Library

30 articles across three knowledge tracks + 3 extended readings:
42 articles across three knowledge tracks + 2 extended readings:

| Track | Coverage | Perspectives |
|-------|----------|-------------|
| AI-Era Harness Engineering | 27 articles | OpenAI → Fowler → Anthropic → LangChain → Stanford → Claude Code reverse engineering → Subagent runtime → Sensors/SPDD/ADLC → Out-of-scope & quality postmortemPractice recap & Engine |
| AI-Era Harness Engineering | 38 articles | OpenAI → Fowler → Anthropic → LangChain → Stanford → Claude Code reverse engineering & source leak → Subagent runtime → Sensors/SPDD/ADLC → Out-of-scope, safety auditing & quality postmortemsEvaluation trilogy → Dynamic workflows → Origins (Ralph / Hashimoto) & discipline synthesis |
| Cloud-Native Harness.io | 2 articles | CI/CD platform architecture (same name, different meaning) |
| Efficiency Paradox & Capability Evolution | 1 article | YDD systematic teardown: TOC + Spec/Rule/Skill |
| Extended Reading | 3 articles | Mitchell Hashimoto, Context Engineering, Human-Agent collaboration |
| Efficiency Paradox & Capability Evolution | 2 articles | YDD systematic teardown + METR follow-up (measurement-methodology crisis) |
| Extended Reading | 2 articles | Context Engineering, Human-Agent collaboration |

See [references/articles.md](references/articles.md) — each article includes core thesis, key data, and cross-article connections.

## 📖 Translations

<details>
<summary><b>22 Chinese translations of key articles</b> (click to expand)</summary>
<summary><b>24 Chinese translations of key articles</b> (click to expand)</summary>

| Translation | Original Author | Source |
|-------------|----------------|--------|
| ⭐ [Eight Years of Wanting](works/maganti-eight-years-building-ai-translation.md) | Lalit Maganti | Personal blog |
| [A Harness for Every Task: Dynamic Workflows](works/anthropic-dynamic-workflows-translation.md) | Thariq Shihipar et al. | Anthropic / Claude |
| [METR: Changing Our Productivity Experiment Design](works/metr-uplift-update-translation.md) | Joel Becker et al. | METR |
| [Inside the Scaffold](works/inside-the-scaffold-paper-translation.md) | Benjamin Rombaut | Huawei / arXiv |
| [Meta-Harness](works/meta-harness-paper-translation.md) | Yoonho Lee et al. | Stanford / arXiv |
| [Harness Engineering (full)](works/fowler-harness-engineering-full-translation.md) | Birgitta Böckeler | Martin Fowler |
Expand Down Expand Up @@ -207,7 +209,7 @@ The "Ralph Wiggum Loop" is the core implementation pattern of Harness Engineerin
| Resource | Description |
|----------|-------------|
| [vibe-coding-cn](https://github.com/tukuaiai/vibe-coding-cn) | Chinese Vibe Coding community guide |
| [Mitchell Hashimoto: Engineer the Harness](https://mitchellh.com/writing/my-ai-adoption-journey#step-5-engineer-the-harness) | Another origin of the "Harness" concept |
| [Mitchell Hashimoto: Engineer the Harness](https://mitchellh.com/writing/my-ai-adoption-journey#step-5-engineer-the-harness) | Where "harness engineering" got its name (now indexed as article #29 in references/articles.md) |

## 🛠️ Development Notes

Expand Down
24 changes: 13 additions & 11 deletions README.md
Original file line number Diff line number Diff line change
@@ -1,8 +1,8 @@
中文 | [English](README.en.md)

![License: MIT](https://img.shields.io/badge/license-MIT-blue)
![Articles](https://img.shields.io/badge/articles-30-green)
![Translations](https://img.shields.io/badge/translations-22-orange)
![Articles](https://img.shields.io/badge/articles-42-green)
![Translations](https://img.shields.io/badge/translations-24-orange)

# Harness Engineering 学习指南

Expand Down Expand Up @@ -111,10 +111,10 @@ harness-engineering/
├── thinking/ # Phase 2:独立思考与质疑(9 篇)
├── practice/ # Phase 3:小项目实验(1 个 Ralph Demo)
├── feedback/ # Phase 4:踩坑与迭代心得(1 篇)
├── works/ # Phase 5:可展示的作品(22 篇翻译 + 1 篇原创 + 2 篇外部中文收录)
├── works/ # Phase 5:可展示的作品(24 篇翻译 + 1 篇原创 + 2 篇外部中文收录)
├── tools/ # 工具具像化:降低 6 维复杂度的杠杆库
├── prompts/ # 验证有效的提示词积累
└── references/ # 外部资源索引(30 篇文章深度摘要)
└── references/ # 外部资源索引(42 篇文章深度摘要)
```

每个子目录都有自己的 `AGENTS.md`,说明该目录的用途和写作约定。这本身就是原文「渐进式披露」的实践。
Expand All @@ -125,29 +125,31 @@ harness-engineering/
- [x] **Phase 2:形成自己的观点** — 9 篇独立思考(持续中)
- [x] **Phase 3:选一个小项目实践** — Ralph Demo 完成(321 秒,$0.31)
- [x] **Phase 4:记录反馈迭代** — 1 篇(持续中)
- [x] **Phase 5:输出可展示的作品** — 22 篇专业翻译 + 1 篇原创综合分析 + 2 篇外部中文收录
- [x] **Phase 5:输出可展示的作品** — 24 篇专业翻译 + 1 篇原创综合分析 + 2 篇外部中文收录

## 📚 研究资料库

跨三条知识脉络 30 篇文章 + 3 篇延伸阅读:
跨三条知识脉络 42 篇文章 + 2 篇延伸阅读:

| 脉络 | 覆盖 | 核心视角 |
|------|------|---------|
| AI 时代的 Harness Engineering | 27 篇 | OpenAI → Fowler → Anthropic → LangChain → Stanford → Claude Code 逆向 → Subagent runtime → 传感器/SPDD/ADLC → 越界与质量复盘实践复盘与 Engine |
| AI 时代的 Harness Engineering | 38 篇 | OpenAI → Fowler → Anthropic → LangChain → Stanford → Claude Code 逆向与源码实锤 → Subagent runtime → 传感器/SPDD/ADLC → 越界·安全审计·质量复盘评测三部曲 → 动态工作流 → 起源考据(Ralph / Hashimoto)与学科汇流 |
| 云原生 Harness.io | 2 篇 | CI/CD 平台架构(同名不同义的参照) |
| 效率悖论与能力进化 | 1 篇 | YDD 系统性拆解:约束理论 + Spec/Rule/Skill |
| 延伸阅读 | 3 篇 | Mitchell Hashimoto、Context Engineering、人机协作 |
| 效率悖论与能力进化 | 2 篇 | YDD 系统性拆解 + METR 实验后续(测量方法论危机) |
| 延伸阅读 | 2 篇 | Context Engineering、人机协作 |

详见 [references/articles.md](references/articles.md) — 每篇文章含核心论点、关键数据、跨文章关联的深度摘要。

## 📖 翻译作品

<details>
<summary><b>22 篇核心文章的中文翻译</b>(点击展开)</summary>
<summary><b>24 篇核心文章的中文翻译</b>(点击展开)</summary>

| 作品 | 原作者 | 来源 |
|------|--------|------|
| ⭐ [渴望了八年,用 AI 三个月造出来](works/maganti-eight-years-building-ai-translation.md) | Lalit Maganti | 个人博客 |
| [为每个任务配一套 harness:动态工作流](works/anthropic-dynamic-workflows-translation.md) | Thariq Shihipar 等 | Anthropic / Claude |
| [METR:我们正在更改生产力实验设计](works/metr-uplift-update-translation.md) | Joel Becker 等 | METR |
| [Inside the Scaffold 论文](works/inside-the-scaffold-paper-translation.md) | Benjamin Rombaut | Huawei / arXiv |
| [Meta-Harness 论文](works/meta-harness-paper-translation.md) | Yoonho Lee 等 | Stanford / arXiv |
| [Harness Engineering 正式版](works/fowler-harness-engineering-full-translation.md) | Birgitta Böckeler | Martin Fowler |
Expand Down Expand Up @@ -206,7 +208,7 @@ harness-engineering/
| 资源 | 说明 |
|------|------|
| [vibe-coding-cn](https://github.com/tukuaiai/vibe-coding-cn) | 中文 Vibe Coding 社区指南 |
| [Mitchell Hashimoto: Engineer the Harness](https://mitchellh.com/writing/my-ai-adoption-journey#step-5-engineer-the-harness) | "Harness" 概念的另一个起源 |
| [Mitchell Hashimoto: Engineer the Harness](https://mitchellh.com/writing/my-ai-adoption-journey#step-5-engineer-the-harness) | "harness engineering" 命名出处(已收录为文章 #29,深度摘要见 references/articles.md) |

## 🛠️ 开发须知

Expand Down
30 changes: 23 additions & 7 deletions prompts/deep-research-tracker.md
Original file line number Diff line number Diff line change
Expand Up @@ -61,11 +61,11 @@
> 它必须自包含,因为搜索器无法访问 `references/articles.md`。
>
> **维护纪律:** 当 `references/articles.md` 新增/删除条目时,**同一次提交中**必须同步更新本节。两份内容的口径(脉络划分、篇数、产品/项目清单)应保持完全一致。
> 本节最近一次同步:2026-06-25(与 `articles.md` 当前内容对齐:30 篇文章 + 1 项已跟踪产品)。
> 本节最近一次同步:2026-07-08(与 `articles.md` 当前内容对齐:42 篇文章 + 1 项已跟踪产品)。

**核心文章 30 篇,分布于三条脉络:**
**核心文章 42 篇,分布于三条脉络:**

- **脉络一 — AI 时代 Harness Engineering(27 篇):**
- **脉络一 — AI 时代 Harness Engineering(38 篇):**
- OpenAI "Harness engineering"(原点,2026-02-11)/ "An open-source spec for Codex orchestration: Symphony"(2026-04-27,任务跟踪器作为控制平面)
- Fowler/Böckeler "Harness engineering for coding agent users"(2026-04-02)+ 前传备忘录(2026-02-17)
- LangChain "The Anatomy of an Agent Harness"(2026-03)/ "Continual Learning for AI Agents"(2026-04-05)/ "Agent Evaluation Readiness Checklist"
Expand All @@ -89,8 +89,19 @@
- Overeager Coding Agents 论文(arXiv 2605.18583,越界动作测量 + 提示声明授权反而降低边界推断)
- Chris Parsons "How I Use AI to Code"(2026-05,四要素 Harness + 从批准者到训练者 + 反馈是新瓶颈;含译者注,2026 数据未独立核实)
- LangChain / Palash Shah "How we built LangSmith Engine"(2026-05-19,用智能体改进智能体,trace→轨迹骨架 + screener/investigator 两阶段闭环)
- Geoffrey Huntley "Ralph Wiggum as a software engineer"(2025-07)+ "everything is a ralph loop"(2026-01-17,bash 循环 + 干净上下文 + 背压 + 单体反多智能体论)
- Mitchell Hashimoto "My AI Adoption Journey"(2026-02-05,六步采纳路线 + "harness engineering" 命名出处)
- Claude Code 源码泄漏事件(2026-03-31,v2.1.88 npm source map 泄 512K 行 TypeScript;聚合分析 pankaj28843/understanding-claude-code:QueryEngine ~46K 行、60+ 门控工具、KAIROS/AutoDream/Undercover Mode)
- Addy Osmani "Agent Harness Engineering"(2026-04-19,O'Reilly 转载;学科汇流综合 + 约束加减法纪律 + hooks 分界论 + HaaS)
- Thoughtworks / Böckeler & Ford "Exploring AI coding sensors"(2026-05-13,有/无传感器对照实验 + harness 模板展望)
- HarnessAudit 论文(arXiv 2605.14271,harness 安全审计:中途轨迹违规是输出级评估盲区,210 任务基准)
- Harness-Bench 论文(arXiv 2605.27922,配置级 harness 效应:106 任务 / 5194 轨迹 + 执行对齐失败)
- "How good is your harness?"(CTB@ICML 2026,Terminal-Bench 2.0 榜单方差统计归因:harness 效应 ≈ 模型效应且异质)
- Anthropic/Claude "A harness for every task: dynamic workflows in Claude Code"(2026-06-02,模型现场写自己的编排 harness + 对抗验证 + ultracode 触发)
- 马东锡 NLP "Harness 才是产品"(2026-06,Model 在 loop 里 harness 拥有 loop + 六组件 + 症状→组件 debug 表)
- Position 论文 "Coding Benchmarks Are Misaligned with Agentic SE"(arXiv 2606.17799,基准折叠 model/harness/环境的三症状)
- **脉络二 — 云原生 Harness.io(2 篇):** Harness.io 官方全局架构 / Google Cloud 集成场景
- **脉络三 — 效率悖论(1 篇):** YDD/Miss-you "效率悖论的系统性拆解"(2026-03-03)
- **脉络三 — 效率悖论(2 篇):** YDD/Miss-you "效率悖论的系统性拆解"(2026-03-03)/ METR 实验后续 + 自报调查(2026-02-24 + 2026-05-11,"慢 19%"的官方后续:弱证据转向加速 + RCT 方法论危机

**已跟踪的开源项目/产品:**
- Ralph Loop 实验链:snarktank/ralph、ralph-orchestrator、bmad-ralph
Expand Down Expand Up @@ -181,12 +192,14 @@
2. **缺口分析**:这批内容覆盖了我们的哪些知识缺口?还有哪些缺口未被触及?
当前已知缺口:
- 行为 Harness(功能正确性验证)
- Harness 覆盖率评估方法
- 跨模型可移植性
- Harness 覆盖率评估方法(#34/#35/#38 评测三部曲已部分回应,组件级信号供给仍开放)
- 跨模型可移植性(#35 给出首个定量异质性证据,迁移指南仍缺)
- 成本数据
- 中小团队实践案例
- Harness 接口维护成本(追宿主升级在总维护中的占比)
- 控制的"激活策略"(always-on / per-commit / conditional / human-summoned 的分类与适用场景)
- 控制的"激活策略"(always-on / per-commit / conditional / human-summoned 的分类与适用场景;#36 的 workflow/ultracode 触发是一个数据点)
- Harness 安全审计(#33 开辟:中途轨迹违规、多智能体信息流、harness 决定安全上限——后续跟踪该方向的评测与工具)
- 评测方法论(基准如何拆分 model/harness/环境的组件级归因——跟踪 #34/#35/#38 的后续工作)

3. **趋势信号**:这批内容中是否有新的趋势或方向?与我们已有的四个学派(约束派、控制论派、架构派、怀疑派)是否一致?

Expand Down Expand Up @@ -214,6 +227,9 @@
5. https://github.com/trending?since=weekly — AI/agent 相关项目
6. https://simonwillison.net/ — 新文章
7. https://mitchellh.com/writing — 新文章
8. https://claude.com/blog — 新文章(Anthropic 产品侧博客;2026-06 的 dynamic workflows / Managed Agents 两篇重磅均发于此,工程博客不覆盖)
9. https://addyosmani.com/blog/ — 新文章
10. https://ghuntley.com/ — 新文章(Ralph 方法论源头)

匹配关键词:harness, agent, coding agent, context engineering,
AGENTS.md, compaction, multi-agent, sandbox
Expand Down
Loading
Loading