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/,27 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/,30 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
19 changes: 11 additions & 8 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-50-green)
![Translations](https://img.shields.io/badge/translations-27-orange)
![Articles](https://img.shields.io/badge/articles-59-green)
![Translations](https://img.shields.io/badge/translations-30-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 (27 translations + 1 original + 2 external Chinese captures)
├── works/ # Phase 5: Shareable outputs (30 translations + 1 original + 2 external Chinese captures)
├── tools/ # Tools that reduce the 6 complexity dimensions
├── prompts/ # Validated prompts collection
└── references/ # External resource index (50 articles with deep summaries)
└── references/ # External resource index (59 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 @@ -126,15 +126,15 @@ 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** — 27 professional translations + 1 original synthesis + 2 external Chinese captures
- [x] **Phase 5: Produce shareable work** — 30 professional translations + 1 original synthesis + 2 external Chinese captures

## 📚 Research Library

50 articles across three knowledge tracks + 2 extended readings:
59 articles across three knowledge tracks + 2 extended readings:

| Track | Coverage | Perspectives |
|-------|----------|-------------|
| AI-Era Harness Engineering | 46 articles | OpenAI → Fowler → Anthropic → LangChain → Stanford → Claude Code reverse engineering & source leak → Subagent runtime → Sensors/SPDD/ADLC → Out-of-scope, safety auditing & quality postmortems → Evaluation trilogy → Dynamic workflows → Origins (Ralph / Hashimoto) & discipline synthesis → Codex harness anatomy → Loop Engineering trilogy → Self-evolving harnesses & RSI → Formal verification |
| AI-Era Harness Engineering | 55 articles | OpenAI → Fowler → Anthropic → LangChain → Stanford → Claude Code reverse engineering & source leak → Subagent runtime → Sensors/SPDD/ADLC → Out-of-scope, safety auditing & quality postmortems → Evaluation trilogy → Dynamic workflows → Origins (Ralph / Hashimoto) & discipline synthesis → Codex harness anatomy → Loop Engineering trilogy → Self-evolving harnesses & RSI → Formal verification → Multi-agent scaling (Cursor / C compiler) → Official containment & evals methodology → Behavior maps / DSLs / local models / outer-loop accountability |
| Cloud-Native Harness.io | 2 articles | CI/CD platform architecture (same name, different meaning) |
| Efficiency Paradox & Capability Evolution | 2 articles | YDD systematic teardown + METR follow-up (measurement-methodology crisis) |
| Extended Reading | 2 articles | Context Engineering, Human-Agent collaboration |
Expand All @@ -144,11 +144,14 @@ See [references/articles.md](references/articles.md) — each article includes c
## 📖 Translations

<details>
<summary><b>27 Chinese translations of key articles</b> (click to expand)</summary>
<summary><b>30 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 |
| [Building a C Compiler with a Team of Parallel Claudes](works/anthropic-c-compiler-translation.md) | Nicholas Carlini | Anthropic |
| [Scaling Long-Running Autonomous Coding](works/cursor-scaling-agents-translation.md) | Wilson Lin | Cursor |
| [How We Contain Claude Across Products](works/anthropic-how-we-contain-translation.md) | Max McGuinness et al. | Anthropic |
| [Harness Engineering for Self-Improvement](works/weng-harness-self-improvement-translation.md) | Lilian Weng | Lil'Log |
| [Loop Engineering](works/osmani-loop-engineering-translation.md) | Addy Osmani | Personal blog |
| [The Coming Loop](works/ronacher-coming-loop-translation.md) | Armin Ronacher | Personal blog |
Expand Down
19 changes: 11 additions & 8 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-50-green)
![Translations](https://img.shields.io/badge/translations-27-orange)
![Articles](https://img.shields.io/badge/articles-59-green)
![Translations](https://img.shields.io/badge/translations-30-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(27 篇翻译 + 1 篇原创 + 2 篇外部中文收录)
├── works/ # Phase 5:可展示的作品(30 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(50 篇文章深度摘要)
└── references/ # 外部资源索引(59 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 @@ -125,15 +125,15 @@ 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** — 27 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** — 30 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 50 篇文章 + 2 篇延伸阅读:
跨三条知识脉络 59 篇文章 + 2 篇延伸阅读:

| 脉络 | 覆盖 | 核心视角 |
|------|------|---------|
| AI 时代的 Harness Engineering | 46 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 → Codex harness 解剖 → Loop Engineering 三部曲 → 自演化 harness 与 RSI → 形式化验证 |
| AI 时代的 Harness Engineering | 55 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 → Codex harness 解剖 → Loop Engineering 三部曲 → 自演化 harness 与 RSI → 形式化验证 → 多智能体并行规模化(Cursor / C compiler)→ 遏制与评测官方方法论 → 行为地图 / DSL / 本地模型 / 外环问责 |
| 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 | 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 |
Expand All @@ -143,11 +143,14 @@ harness-engineering/
## 📖 翻译作品

<details>
<summary><b>27 篇核心文章的中文翻译</b>(点击展开)</summary>
<summary><b>30 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 |
| [用一支并行 Claude 团队构建 C 编译器](works/anthropic-c-compiler-translation.md) | Nicholas Carlini | Anthropic |
| [规模化长时自主编码](works/cursor-scaling-agents-translation.md) | Wilson Lin | Cursor |
| [我们如何在各产品中遏制 Claude](works/anthropic-how-we-contain-translation.md) | Max McGuinness 等 | Anthropic |
| [面向自我改进的 Harness Engineering](works/weng-harness-self-improvement-translation.md) | Lilian Weng | Lil'Log |
| [循环工程(Loop Engineering)](works/osmani-loop-engineering-translation.md) | Addy Osmani | 个人博客 |
| [正在到来的循环(The Coming Loop)](works/ronacher-coming-loop-translation.md) | Armin Ronacher | 个人博客 |
Expand Down
32 changes: 29 additions & 3 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:2026-07-15(与 `articles.md` 当前内容对齐:50 篇文章 + 1 项已跟踪产品)。
> 本节最近一次同步:2026-07-21(与 `articles.md` 当前内容对齐:59 篇文章 + 1 项已跟踪产品)。

**核心文章 50 篇,分布于三条脉络:**
**核心文章 59 篇,分布于三条脉络:**

- **脉络一 — AI 时代 Harness Engineering(46 篇):**
- **脉络一 — AI 时代 Harness Engineering(55 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 Down Expand Up @@ -108,6 +108,15 @@
- Self-Harness 论文(arXiv 2606.09498,智能体自改 harness:弱点挖掘→提议→回归验证,三模型 held-out +14~21pp)
- Lilian Weng "Harness Engineering for Self-Improvement"(2026-07-04,RSI 综述:三设计模式 + 优化对象递进链 + 七项未来挑战)
- Aria 论文 "Harnessing Code Agents for Automatic Software Verification"(arXiv 2607.06341,声明式 HHL 包裹通用智能体做 Coq 证明,行为 harness 极限形态)
- Anthropic "Demystifying evals for AI agents"(2026-01-09,智能体评测官方方法论:outcome≠transcript、pass@k/pass^k、CORE-Bench 42%→95% 基准病理)
- Cursor / Wilson Lin "Scaling long-running autonomous coding"(2026-01-14,数百并发智能体 × 数周:planner/worker/judge + FastRender 百万行浏览器)
- Anthropic / Nicholas Carlini "Building a C compiler with a team of parallel Claudes"(2026-02-05,16 agent 无编排者 + GCC oracle + $20k 成本账本)
- Anthropic "How we contain Claude across products"(2026-05-25,三隔离模式 + 五起漏掉的风险:白名单是能力授予、先环境层遏制)
- LangChain "Introducing Dynamic Subagents in Deep Agents"(2026-06-29)+ "How to Use RLMs in Deep Agents"(2026-07-01,模型写编排脚本 + RLM 递归上下文)
- Fowler/Böckeler 本地模型双备忘录(2026-07 上旬,Qwen3.6 35B MoE 甜点 + 工具调用是智能体化分水岭)
- Harness Handbook 论文(arXiv 2607.13285,2026-07-14,行为定位是 harness 演化瓶颈:行为中心表示 + BGPD)
- Fowler / Unmesh Joshi "DSLs Enable Reliable Use of LLMs"(2026-07-14,DSL 工具集即 harness:语言层约束 + 确定性验证器)
- Addy Osmani "Own the Outer Loop"(2026-07-15,AIE 闭幕演讲:智能体跑内环、工程师拥有外环问责 + back-pressure 调速)
- **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(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)

Expand Down Expand Up @@ -277,3 +286,20 @@
│ 你决定:收录 / 翻译 / 写分析 / 跳过 │
└─────────────────────────────────────────────────┘
```

## 核心信源存量回扫(每批必做一步)

> 教训来自 2026-07-21 批次:只按"最近 2 周"时间窗扫描,导致 4 篇 1–5 月的重量级官方件
> (Cursor scaling-agents、Anthropic C compiler / how-we-contain / demystifying-evals)漏网数月。
> 时间窗抓增量,回扫补存量——两者缺一不可。

每轮调研除时间窗扫描外,**轮换抽取 1–2 个 Tier 1 信源,翻其全年归档目录**(archive / blog 列表页),
对照 `references/articles.md`(含观察项表)逐条核对是否已收录或已甄别。优先轮换顺序:

1. anthropic.com/engineering(工程博客,全量列表)
2. cursor.com/blog(research 分类)
3. openai.com/index(Engineering 分类)
4. martinfowler.com/tags(GenAI 标签 + exploring-gen-ai 系列)
5. langchain.com/blog、claude.com/blog、addyosmani.com/blog、simonwillison.net 月归档

发现漏网存量时,与本批增量一起进「候选 × 定性 × 去向」评审表,注明"存量补课"。
Loading
Loading