Skip to content

Commit 9f8eb00

Browse files
wanshuiyinwanshuiyin
committed
feat: CONTRIBUTING + bundled skill source + landing HTML + Agent Foundations
Repo substance / scaffolding: - CONTRIBUTING.md + CONTRIBUTING_CN.md — bilingual contribution guide with strict style rules baked in (heading conv, math escapes, callout-list spacing, personal-info banlist) and PR checklist - skills/render-html/SKILL.md + skills/interview-cheatsheet/SKILL.md — skill source bundled (mirror from ARIS main repo) so contributors can fork the workflow directly - tools/render_html.py + tools/templates/{academic,dashboard}.html — the actual renderer used to produce every HTML in docs/tutorials/ (gives the repo real Python code, not just docs) - docs/about.md → docs/index.html — landing page at the Pages root (https://wanshuiyin.github.io/ARIS-in-AI-Offer/), with workflow diagram, audit-trail case studies, and category overview New tutorial: - 🤖 Agent Foundations (1229 lines, 9 codex review rounds catching ReAct table numbers, SWE-bench Verified OpenAI attribution, MCP 2025-11-25 DCR demoted to MAY, A2A v1.0 SCREAMING_SNAKE_CASE, Anthropic Tool Use GA 2024-05-30, Agent S3 three-tier 62.6/69.9/72.6, and ~25 more) Both READMEs (CN + EN): - New 🤖 Agents category with Agent Foundations row promoted + 3 🚧 - Footer updated: 16/19 tutorials live, 3 Agents being DIY-reviewed - Contributing section now points to CONTRIBUTING.md
1 parent 635ebc7 commit 9f8eb00

14 files changed

Lines changed: 5303 additions & 5 deletions

‎CONTRIBUTING.md‎

Lines changed: 80 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,80 @@
1+
# Contributing to ARIS-in-AI-Offer
2+
3+
[English](CONTRIBUTING.md) | [中文版](CONTRIBUTING_CN.md)
4+
5+
Thanks for stopping by 🌱 — this repo exists so that 2026 AI campus recruiting (秋招) is a bit less painful, and so future cohorts have a base to build on. One person can only cover so much; PRs and issues are equally welcome.
6+
7+
## Ways to contribute
8+
9+
- **Add a new cheat sheet** on a topic the current collection doesn't cover (e.g. Diffusion Post-Training, Flow-DPO/OPD, Audio Gen, LLM-as-Judge, Long-Video Diffusion, …).
10+
- **Update an existing cheat sheet** when new frontier work breaks an old claim (e.g. new SOTA on SWE-bench, a new MCP spec revision).
11+
- **Fix a math / code / citation bug** — every tutorial has a cross-model `codex GPT-5.5 xhigh` audit trail in `*.review.json`, but reviewers miss things too.
12+
- **Improve rendering / styling** — `tools/render_html.py` and `tools/templates/` live in this repo; PR rendering improvements there.
13+
- **Translate** — add an English version of a tutorial body (currently most tutorials are Chinese with English technical terms inline).
14+
15+
## How to add a new cheat sheet — the ARIS workflow
16+
17+
Every tutorial in this repo was generated by ARIS's `/interview-cheatsheet` + `/render-html` skills. To stay consistent, **please use the same workflow** when you contribute:
18+
19+
1. **Install ARIS** (skill source is bundled in `skills/`; canonical home is the [ARIS main repo](https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep)). With Claude Code: `bash <(curl -s https://raw.githubusercontent.com/wanshuiyin/Auto-claude-code-research-in-sleep/main/tools/install_aris.sh)`
20+
2. **Invoke** `/interview-cheatsheet "<TOPIC>"` from inside this repo. The skill auto-runs the cross-model codex review gate + render-html.
21+
3. **Review** the generated `.md` / `.html` / `.review.json` files yourself before pushing.
22+
4. **Open a PR** with all three files (MD source, HTML render, review JSON) plus a row added to README.md / README_EN.md in the right category.
23+
24+
If you want to do it by hand without ARIS, that's fine too — just match the structure of an existing tutorial (e.g. `docs/tutorials/attention_tutorial.md`) and run `tools/render_html.py` to render.
25+
26+
## Style guide (strict)
27+
28+
These rules came from review-loop lessons learned across the first 16 tutorials:
29+
30+
| Rule | Why | How |
31+
|---|---|---|
32+
| Heading: `## §N Title` (space after §N) | Older `§NTitle` glued | always have a space |
33+
| Math in tables: `\lvert x \rvert` not `\|x\|` | `\|` in markdown table = cell separator | use LaTeX explicit form |
34+
| `<` in math: `\lt`, `>` in math: `\gt` | Raw `<t` gets parsed as HTML start tag, breaks render | escape via `\lt` / `\gt` |
35+
| Callouts (`> 💡 / ⚠️ / ✅ / ❌`) followed by list: **blank line between** | Else renderer ambiguity | enforce blank-line separator |
36+
| Sections: §0 TL;DR · §1 directly-grokable intuition · §2-§9 math+code+variants · §10 25-Q&A (L1 必会 × 10 / L2 进阶 × 10 / L3 顶级 lab × 5) · §A optional appendix | matches ARIS pilot pattern | follow numbered structure |
37+
| No personal info | privacy + general portability | byline goes via `--author` flag at render time; ban: SJTU JHC / Server5 / job market / `/Users/...` / specific lab |
38+
39+
See `skills/render-html/SKILL.md` and `skills/interview-cheatsheet/SKILL.md` for the full contract.
40+
41+
## Pull request process
42+
43+
1. Make sure your changes are well-documented (include the `.review.json` so reviewers can see your audit trail).
44+
2. Update **both** `README.md` and `README_EN.md` if you add a new tutorial — pick the right category bucket (General / Post-Training & Reasoning / LLM Architecture & Systems / Generative Models — Theory & Tokenizers / Generation Systems / Multimodal / Agents).
45+
3. Keep PRs focused on a single topic — one tutorial per PR.
46+
4. Clear commit messages — `docs(tutorials): add <Topic> cheat sheet (rendered via /render-html)` is the convention.
47+
48+
### PR checklist
49+
50+
- [ ] `.md` source follows the style guide above
51+
- [ ] `.html` is freshly rendered via `tools/render_html.py`
52+
- [ ] `.review.json` includes cross-model codex audit trail (math/code review verdict + render review verdict + thread IDs)
53+
- [ ] No personal info beyond the byline
54+
- [ ] Both `README.md` + `README_EN.md` updated
55+
- [ ] Citations are real and verifiable (no hallucinated authors / years / venues)
56+
57+
## Honest disclaimer
58+
59+
The existing tutorials are best-effort but the field moves fast. If you spot:
60+
- a paper credited to the wrong author / year / venue,
61+
- a math derivation that drops a term,
62+
- code that wouldn't actually run,
63+
- a claim about a benchmark / SOTA that's now stale,
64+
- or just unclear writing,
65+
66+
… please open an issue or PR. We'll fix it together. Reviewers (including cross-model codex audits) miss things; community eyes catch them.
67+
68+
## Code of conduct
69+
70+
- Be respectful and constructive — assume good faith.
71+
- Focus feedback on the work, not the person.
72+
- Help others learn; this is a study resource first.
73+
74+
## Questions?
75+
76+
Open an issue, or come find us in the [ARIS main repo](https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep) WeChat / Discord (QR + link in that repo's README).
77+
78+
## License
79+
80+
By contributing, you agree your contributions will be licensed under MIT — same as the rest of the repo.

‎CONTRIBUTING_CN.md‎

Lines changed: 80 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,80 @@
1+
# 给 ARIS-in-AI-Offer 贡献
2+
3+
[English](CONTRIBUTING.md) | 中文版
4+
5+
谢谢你愿意看到这里 🌱 —— 这个仓库存在的初衷就是让 2026 年 AI 秋招稍微轻松一点，也给以后的同学留个底子。一个人的力量有限，PR 和 issue 都欢迎。
6+
7+
## 你可以怎么贡献
8+
9+
- **加一篇新 cheat sheet**：当前 collection 没覆盖的方向（Diffusion Post-Training、Flow-DPO/OPD、Audio Gen、LLM-as-Judge、Long-Video Diffusion …）
10+
- **更新现有 cheat sheet**：前沿动了（SWE-bench 新 SOTA、MCP spec 新 revision 等）
11+
- **修数学 / 代码 / 引用 bug**：每篇都有跨模型 `codex GPT-5.5 xhigh` 审查 trail 在 `*.review.json` 里，但 reviewer 也会漏
12+
- **改进 render / 排版**：`tools/render_html.py` 和 `tools/templates/` 都在这个仓库里，直接 PR
13+
- **翻译**：当前大部分教程是中文 + 英文术语原文混排，可以加纯英文版
14+
15+
## 怎么加新 cheat sheet —— ARIS workflow
16+
17+
仓库里所有教程都是用 ARIS 的 `/interview-cheatsheet` + `/render-html` 两个 skill 生成的。为了保持一致，**请也用这套 workflow**：
18+
19+
1. **安装 ARIS**（skill 源码在 `skills/`；canonical 来源是 [ARIS 主仓库](https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep)）。Claude Code 用户：`bash <(curl -s https://raw.githubusercontent.com/wanshuiyin/Auto-claude-code-research-in-sleep/main/tools/install_aris.sh)`
20+
2. **在本仓库目录里调** `/interview-cheatsheet "<TOPIC>"`。skill 自动跑跨模型 codex review gate + render-html。
21+
3. **自己 review** 生成的 `.md` / `.html` / `.review.json` 再 push。
22+
4. **开 PR**：三个文件都带上（MD 源 / HTML 渲染 / review JSON），README.md / README_EN.md 对应分类下加一行。
23+
24+
不想用 ARIS 手写也行 —— 照着现有教程的结构写（例如 `docs/tutorials/attention_tutorial.md`），跑 `tools/render_html.py` 渲染就行。
25+
26+
## 风格指南（严格）
27+
28+
这些规则是前 16 篇 review-loop 踩出来的：
29+
30+
| 规则 | 原因 | 怎么办 |
31+
|---|---|---|
32+
| 标题：`## §N Title`（§N 后面留空格） | 老版本 `§NTitle` 粘连 | 始终留空格 |
33+
| 表格里的数学：`\lvert x \rvert` 不要 `\|x\|` | 表格里的 `\|` 会被当作单元格分隔 | 用 LaTeX 显式写法 |
34+
| 数学里的 `<` 用 `\lt`，`>` 用 `\gt` | 原生 `<t` 会被 parser 当 HTML 起始 tag，整段坏 | 用 `\lt` / `\gt` 转义 |
35+
| Callout (`> 💡 / ⚠️ / ✅ / ❌`) 后接 list：**中间留空行** | 否则 renderer 会把 list 吃进 callout | 强制空行分隔 |
36+
| 章节：§0 TL;DR · §1 直觉 · §2-§9 公式+代码+变体 · §10 25 高频题（L1 必会 × 10 / L2 进阶 × 10 / L3 顶级 lab × 5） · §A 可选附录 | 匹配 ARIS pilot pattern | 按编号结构走 |
37+
| 不放个人信息 | 隐私 + 通用性 | byline 走 `--author` flag 在 render 时加；ban：SJTU JHC / Server5 / job market / `/Users/...` / 具体实验室 |
38+
39+
完整契约见 `skills/render-html/SKILL.md` 和 `skills/interview-cheatsheet/SKILL.md`。
40+
41+
## PR 流程
42+
43+
1. 改动要文档化（带上 `.review.json` 让 reviewer 看你的 audit trail）
44+
2. 加新教程要 **同时** 更新 `README.md` 和 `README_EN.md`，按分类放（General / Post-Training & Reasoning / LLM Architecture & Systems / Generative Models — Theory & Tokenizers / Generation Systems / Multimodal / Agents）
45+
3. PR 聚焦单一主题 —— 一篇教程一个 PR
46+
4. Commit 信息清晰 —— 规范是 `docs(tutorials): add <Topic> cheat sheet (rendered via /render-html)`
47+
48+
### PR checklist
49+
50+
- [ ] `.md` 源符合风格指南
51+
- [ ] `.html` 是用 `tools/render_html.py` 重新渲染的
52+
- [ ] `.review.json` 包含跨模型 codex audit trail（math/code review verdict + render review verdict + thread IDs）
53+
- [ ] byline 之外没个人信息
54+
- [ ] `README.md` + `README_EN.md` 都更新了
55+
- [ ] 引用真实可验证（不能 hallucinate 作者 / 年份 / venue）
56+
57+
## 坦白说
58+
59+
现有教程是尽力而为，但领域更新太快。如果你发现：
60+
- 作者 / 年份 / venue 标错了
61+
- 数学推导漏一项
62+
- 代码实际跑不通
63+
- benchmark / SOTA 已经过时
64+
- 或者就是写得不清楚
65+
66+
…… 请提 issue 或 PR，一起修。Reviewer（含 codex 跨模型审）也会漏，社区的眼睛能补上。
67+
68+
## 行为准则
69+
70+
- 礼貌 + 建设性 —— 假设对方善意
71+
- 反馈对事不对人
72+
- 帮别人学 —— 这首先是个学习资源
73+
74+
## 有问题？
75+
76+
开 issue，或者去 [ARIS 主仓库](https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep) 找微信/Discord 群（二维码 + 链接在那个仓库的 README）。
77+
78+
## License
79+
80+
提交贡献即默认你接受 MIT license —— 和仓库其它部分一致。

‎README.md‎

Lines changed: 14 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -91,7 +91,16 @@
9191
|---|---|---|
9292
| **VLM (CLIP / LLaVA / Qwen-VL / DeepSeek-VL)** | [📄 HTML](https://wanshuiyin.github.io/ARIS-in-AI-Offer/tutorials/vlm_multimodal_tutorial.html) | [MD](docs/tutorials/vlm_multimodal_tutorial.md) |
9393

94-
> ✅ **15 篇齐了** —— 第一波全部就位（2026-05）。后续如果有新方向（Diffusion Post-Training、Flow-OPD、Audio Gen、新 SOTA 方法等）会逐步加；欢迎 PR。
94+
### 🤖 Agents
95+
96+
| Topic | HTML | MD |
97+
|---|---|---|
98+
| **Agent Foundations (ReAct / MCP / A2A / SWE-bench / GAIA / OSWorld)** | [📄 HTML](https://wanshuiyin.github.io/ARIS-in-AI-Offer/tutorials/agent_foundations_tutorial.html) | [MD](docs/tutorials/agent_foundations_tutorial.md) |
99+
| Agentic RL (RL for tool use / agent trajectories) | 🚧 | — |
100+
| Multi-Agent & Long-Horizon (CAMEL / AutoGen / MetaGPT / MemGPT) | 🚧 | — |
101+
| Self-Evolving Agents (Ctx2Skill / Native Evolution / Voyager / Reflexion) | 🚧 | — |
102+
103+
> 🚀 **16 篇就位**（含 Agent Foundations，9 轮 codex 严格审过）；剩 3 篇 Agents 在主对话 DIY 中。后续 Diffusion Post-Training / Flow-OPD / Audio Gen / 更新 SOTA 等会逐步加 —— PR 欢迎。
95104
96105
---
97106

@@ -114,9 +123,11 @@
114123

115124
一个人的力量有限，希望靠大家一起把这套教程做得更完善。
116125

117-
如果你想加一篇 cheat sheet 或修正现有内容，欢迎用 ARIS 的 [`/interview-cheatsheet`](https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep/blob/main/skills/interview-cheatsheet/SKILL.md) + [`/render-html`](https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep/blob/main/skills/render-html/SKILL.md) workflow 跑出来再提 PR。两个 skill 都内置跨模型 codex 5.5 xhigh 审查 gate（数学 / 代码 / 引用 / 渲染保真），过了 PR 进来就有质量底线。
126+
完整贡献指南见 [**CONTRIBUTING.md**](CONTRIBUTING.md)（[English](CONTRIBUTING.md) · [中文](CONTRIBUTING_CN.md)）—— 含 ARIS workflow 调用、严格风格指南（headings / math / tables / callouts / 个人信息 banlist）、PR checklist。
127+
128+
**TL;DR**：用 ARIS 的 [`/interview-cheatsheet`](skills/interview-cheatsheet/SKILL.md) + [`/render-html`](skills/render-html/SKILL.md) workflow 跑出来再提 PR；两个 skill 都内置跨模型 codex 5.5 xhigh 审查 gate（数学 / 代码 / 引用 / 渲染保真），过了就有质量底线。Skill 源码和 `tools/render_html.py` 都在这个仓库里，可以直接 fork。
118129

119-
**坦白说**：现有教程把 HTML 基础结构（公式 / 代码 / 表格 / callout / TOC / 响应式）都做扎实了，但具体到某些主题最前沿（比如 2025 下半年才出的方法、某些细分领域的最新论文）大概率没全覆盖。发现哪里过时或有错，PR / issue 都欢迎，一起把这份资料持续维护。
130+
**坦白说**：现有教程把 HTML 基础结构（公式 / 代码 / 表格 / callout / TOC / 响应式）做扎实了，但某些主题最前沿（2025 下半年才出的方法、某些细分领域最新论文）大概率没全覆盖。发现哪里过时或有错，PR / issue 都欢迎。
120131

121132
---
122133

‎README_EN.md‎

Lines changed: 13 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -91,7 +91,16 @@ Core methodology: **cross-model adversarial review** — executor and reviewer m
9191
|---|---|---|
9292
| **VLM (CLIP / LLaVA / Qwen-VL / DeepSeek-VL)** | [📄 HTML](https://wanshuiyin.github.io/ARIS-in-AI-Offer/tutorials/vlm_multimodal_tutorial.html) | [MD](docs/tutorials/vlm_multimodal_tutorial.md) |
9393

94-
> ✅ **15 tutorials live** — initial batch complete (2026-05). New topics (e.g. Diffusion Post-Training, Flow-OPD, Audio Gen, fresh SOTA methods) will be added over time; PRs welcome.
94+
### 🤖 Agents
95+
96+
| Topic | HTML | MD |
97+
|---|---|---|
98+
| **Agent Foundations (ReAct / MCP / A2A / SWE-bench / GAIA / OSWorld)** | [📄 HTML](https://wanshuiyin.github.io/ARIS-in-AI-Offer/tutorials/agent_foundations_tutorial.html) | [MD](docs/tutorials/agent_foundations_tutorial.md) |
99+
| Agentic RL (RL for tool use / agent trajectories) | 🚧 | — |
100+
| Multi-Agent & Long-Horizon (CAMEL / AutoGen / MetaGPT / MemGPT) | 🚧 | — |
101+
| Self-Evolving Agents (Ctx2Skill / Native Evolution / Voyager / Reflexion) | 🚧 | — |
102+
103+
> 🚀 **16 tutorials live** (including Agent Foundations, audited across 9 codex rounds); 3 more Agents being DIY-reviewed in the main session. Diffusion Post-Training / Flow-OPD / Audio Gen / further SOTA updates will be added over time — PRs welcome.
95104
96105
---
97106

@@ -114,7 +123,9 @@ Cross-model adversarial review (executor ≠ reviewer family) is ARIS's core inv
114123

115124
One person can only cover so much. The hope is that many hands make this collection more complete.
116125

117-
If you'd like to add a cheat sheet or fix existing content, please use ARIS's [`/interview-cheatsheet`](https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep/blob/main/skills/interview-cheatsheet/SKILL.md) + [`/render-html`](https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep/blob/main/skills/render-html/SKILL.md) workflow to generate it, then open a PR here. Both skills enforce a cross-model codex GPT-5.5 xhigh review gate (math / code / citations / render fidelity), so anything merged through PR has a baseline quality floor.
126+
Full contribution guide: [**CONTRIBUTING.md**](CONTRIBUTING.md) ([English](CONTRIBUTING.md) · [中文](CONTRIBUTING_CN.md)) — covers ARIS workflow invocation, strict style guide (headings / math / tables / callouts / personal-info banlist), and PR checklist.
127+
128+
**TL;DR**: use the [`/interview-cheatsheet`](skills/interview-cheatsheet/SKILL.md) + [`/render-html`](skills/render-html/SKILL.md) workflow to generate, then open a PR. Both skills enforce a cross-model codex GPT-5.5 xhigh review gate (math / code / citation / render fidelity), so anything merged via PR has a baseline quality floor. Skill source and `tools/render_html.py` are bundled in this repo so you can fork & extend.
118129

119130
**Honest disclaimer**: across the existing tutorials, the HTML structural foundations (math, code, tables, callouts, TOC, responsive layout) are solid. But the very latest frontier work in any given topic (e.g., methods released in late 2025, niche subfield updates) likely is not fully covered. If you spot something outdated or wrong, PRs and issues are equally welcome — let's keep this resource alive together.
120131

0 commit comments

Comments
 (0)