Skip to content

Commit 6bdb38e

Browse files
yanglbmeyanglbme
committed
docs: add AGENTS.md with formatting and safety guidelines
1 parent 01de3cb commit 6bdb38e

5 files changed

Lines changed: 158 additions & 24 deletions

File tree

‎AGENTS.md‎

Lines changed: 142 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,142 @@
1+
# 🤖 Awesome AI Agent Guide
2+
3+
本文档旨在指导 AI Agent 如何规范、安全地为 **Awesome AI** 项目生成内容、执行更新和部署。
4+
5+
## 📋 项目概览
6+
- **仓库**: `doocs/awesome-ai`
7+
- **线上地址**: https://ai.doocs.org
8+
- **技术栈**: Astro + Tailwind CSS
9+
- **核心策略**: **严格双语同步** (中文/英文)，所有核心内容更新必须包含双语版本。
10+
11+
## 📁 项目结构
12+
```text
13+
awesome-ai/
14+
├── src/
15+
│ ├── content/ # 📂 中文内容源文件
16+
│ │ ├── daily/ # 日报 (文件名: YYYY-MM-DD.md)
17+
│ │ ├── weekly/ # 周报 (文件名: YYYY-wWW.md)
18+
│ │ ├── topics/ # 专题深度讨论
19+
│ │ ├── tools/ # 工具推荐分类
20+
│ │ └── models/ # 模型排行榜分类
21+
│ ├── content/en/ # 📂 英文内容源文件 (结构对应中文)
22+
│ ├── pages/ # Astro 路由页面
23+
│ ├── components/ # UI 组件
24+
│ └── i18n/ # 国际化翻译字典 (zh.json / en.json)
25+
├── public/ # 📂 静态资源 (⚠️ 包含关键文件: CNAME)
26+
├── scripts/ # 🔧 部署脚本 (deploy_ai_pulse.py)
27+
├── README.md # 📖 项目介绍 (中文)
28+
└── README_EN.md # 📖 项目介绍 (英文)
29+
```
30+
31+
## 📝 内容生成规范 (CRITICAL)
32+
33+
### 1. Frontmatter 格式红线 ⚠️
34+
所有 Markdown 文件的 Frontmatter **必须**严格符合 `src/content/config.ts` 中的 Schema 定义。
35+
36+
- **致命错误**：`date` 字段**必须加双引号**！
37+
- ✅ `date: "2026-05-11"` (正确)
38+
- ❌ `date: 2026-05-11` (会导致 Astro 构建失败)
39+
40+
**标准模板：**
41+
```yaml
42+
---
43+
title: "Awesome AI 日报 | 2026-05-11"
44+
date: "2026-05-11"
45+
tags: ["Tag1", "Tag2"]
46+
summary: "简短总结..."
47+
---
48+
```
49+
50+
### 2. 标题与排版统一规范 ⭐️
51+
所有生成的 Markdown 文件在标题层级（Heading levels）上必须保持高度统一。
52+
53+
**📄 日报 (Daily) 模板:**
54+
- 不使用 `## 今日要闻` 等包裹标题，直接以 `## 1.` 开始。
55+
- 每条新闻使用 `##` (H2) 作为标题。
56+
- 观点部分使用引用块 `> **Awesome AI 观点：**`。
57+
```markdown
58+
## 1. 新闻标题
59+
新闻内容段落...
60+
> **Awesome AI 观点：** 观点内容...
61+
62+
## 2. 新闻标题
63+
新闻内容段落...
64+
65+
## 其他动态
66+
- **公司/人物** 动态详情
67+
```
68+
69+
**📰 周报 (Weekly) 模板:**
70+
- 开头使用一个 `#` (H1) 作为本期主标题。
71+
- 每个独立板块使用 `##` (H2)。
72+
```markdown
73+
# 本周聚焦：核心主题一句话概括
74+
75+
简介段落...
76+
77+
## 板块一标题
78+
内容...
79+
80+
## 板块二标题
81+
内容...
82+
```
83+
84+
**💡 专题 (Topics) 模板:**
85+
- 不使用 `#` (H1)，所有内容区块使用 `##` (H2)。
86+
```markdown
87+
> 引导语/核心观点引用
88+
89+
## 板块一标题
90+
内容...
91+
```
92+
93+
### 3. 内容质量
94+
- **观点输出**：每条资讯/工具推荐后需包含 `> **Awesome AI 观点/View:**`，提供深度分析，拒绝简单堆砌。
95+
- **双语一致**：英文版内容需适配英文语境，保持专业度。
96+
97+
## 🔄 更新与部署流程
98+
99+
### Step 1: 生成与验证
100+
1. 在 `src/content/` 和 `src/content/en/` 生成对应的双语 Markdown 文件。
101+
2. 运行本地验证：`npm run build`。**只有构建成功 (Exit 0) 的文件才允许提交。**
102+
103+
### Step 2: 评估并同步 README (必须执行) ⭐️
104+
**规则**：在更新内容后，Agent **必须**评估本次内容的重要性，决定是否更新根目录的 `README.md` 和 `README_EN.md`。
105+
106+
- **何时更新 README？**
107+
- 新增了 **Tools** 分类或 **Tools** 推荐。
108+
- 新增了 **Models** 排行榜或模型数据。
109+
- 新增了高价值的 **Topics**（专题讨论）。
110+
- 项目结构发生重大变化（新增模块）。
111+
- **操作要求**：
112+
- 若评估需要更新，**必须同时修改** 中文版 (`README.md`) 和 英文版 (`README_EN.md`)。
113+
- 保持中英文描述的准确性和一致性。
114+
115+
### Step 3: 部署推送
116+
使用 `scripts/deploy_ai_pulse.py` 推送文件。
117+
```bash
118+
# 推送中文
119+
python3 scripts/deploy_ai_pulse.py src/content/daily/YYYY-MM-DD.md "docs: add daily news YYYY-MM-DD"
120+
121+
# 推送英文
122+
python3 scripts/deploy_ai_pulse.py src/content/en/daily/YYYY-MM-DD.md "docs: add daily news EN YYYY-MM-DD"
123+
```
124+
125+
## 🔒 安全与敏感信息规范
126+
1. **Token 保密**：
127+
- 部署脚本 (`deploy_ai_pulse.py`) 会从环境文件读取 Token。**Agent 在执行脚本时，绝不能在日志、回复或输出中打印 Token 的值。**
128+
- 若遇到认证失败，仅提示 "Authentication failed" 或 "Check token validity"。
129+
2. **禁止硬编码**：
130+
- 永远不要将 Token 硬编码在生成的脚本代码或 Markdown 内容中。
131+
- 仓库根目录或 `public/` 下禁止出现 `.env` 或包含密钥的临时文件。
132+
133+
## 🚀 Git Commit 规范
134+
遵循 [Conventional Commits](https://www.conventionalcommits.org/)：
135+
136+
| Type | 场景 | 示例 |
137+
|:---|:---|:---|
138+
| `docs` | 日报、周报、专题文章、README | `docs: add daily news 2026-05-11` |
139+
| `feat` | 新功能 (页面、组件) | `feat: add dark mode toggle` |
140+
| `fix` | 修复错误 (格式、链接、代码) | `fix: correct date format in daily news` |
141+
| `chore` | 构建配置、CNAME、依赖 | `chore: update CNAME to ai.doocs.org` |
142+
| `refactor` | 代码重构 (无功能变动) | `refactor: optimize image loading` |

‎src/content/daily/2026-05-09.md‎

Lines changed: 4 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -5,27 +5,25 @@ tags: ["Nvidia", "DeepSeek", "Anthropic", "OpenAI", "百度"]
55
summary: "Nvidia今年已投400亿美元AI股权交易；DeepSeek完成500亿融资V4.1定档6月；百度文心5.1发布；Cloudflare称AI淘汰了1100个岗位"
66
---
77

8-
## 今日要闻
9-
10-
### 1. Nvidia 今年已砸下 400 亿美元 AI 股权投资
8+
## 1. Nvidia 今年已砸下 400 亿美元 AI 股权投资
119

1210
Nvidia 在 AI 基础设施上的投资节奏越来越激进。TechCrunch 报道，仅今年前几个月，Nvidia 就已经在 AI 股权交易中投入了 400 亿美元。这个数字比很多人预想的还要夸张——黄仁勋不只是卖芯片，还在用资本把整个 AI 生态绑在自己的战车上。
1311

1412
> 来源：TechCrunch
1513
16-
### 2. DeepSeek 创纪录融资 500 亿，V4.1 定档 6 月
14+
## 2. DeepSeek 创纪录融资 500 亿，V4.1 定档 6 月
1715

1816
量子位报道，梁文锋出资 200 亿领投，DeepSeek 完成了首轮 500 亿人民币的融资，创下了国内 AI 公司的纪录。V4.1 版本已经定档 6 月发布。去年 DeepSeek 靠开源模型在国际上杀出一条路，现在有钱有技术，接下来就看它怎么跟闭源阵营硬碰硬了。
1917

2018
> 来源：量子位
2119
22-
### 3. 百度发布文心 5.1：搜索登顶国内，训练成本只有业界 6%
20+
## 3. 百度发布文心 5.1：搜索登顶国内，训练成本只有业界 6%
2321

2422
百度今天发布了文心 5.1，搜索能力登顶国内所有大模型。更值得关注的是预训练成本——百度声称只有业界的 6%。如果这个数字属实，意味着百度的底层训练效率确实做了大幅优化。在大模型军备竞赛进入"拼内功"阶段的今天，成本控制可能比参数规模更有说服力。
2523

2624
> 来源：量子位
2725
28-
### 4. Cloudflare 说 AI 让 1,100 个岗位过时了，但收入创了历史新高
26+
## 4. Cloudflare 说 AI 让 1,100 个岗位过时了，但收入创了历史新高
2927

3028
一边是收入创新高，一边是 AI 替代了 1,100 个岗位。Cloudflare 的这个表态很有代表性——AI 对就业的冲击已经不再是预测，而是正在发生的现实。有趣的是，公司并没有掩饰这件事，反而把它和业绩增长放在一起说。
3129

‎src/content/daily/2026-05-10.md‎

Lines changed: 4 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -5,27 +5,25 @@ tags: ["Nvidia", "AGI", "AI监管", "OpenAI", "Google AI"]
55
summary: "Nvidia今年已在AI股权投资上砸了400亿美元；AGI概念遭遇集体重包装；Google宣布在AI搜索概览中加入更多来源链接"
66
---
77

8-
## 今日要闻
9-
10-
### 1. Nvidia今年AI股权投资已达400亿美元
8+
## 1. Nvidia今年AI股权投资已达400亿美元
119

1210
Nvidia在AI领域的投资步伐越来越猛。今年才过去不到五个月，这家芯片巨头已经承诺了400亿美元的股权AI交易。这说明什么？Nvidia不仅靠卖芯片赚钱，它正在用资本把自己嵌进整个AI产业链。被投公司越多，护城河就越深。
1311

1412
> 来源：TechCrunch
1513
16-
### 2. "AGI"这个词，正在被AI公司集体重新包装
14+
## 2. "AGI"这个词，正在被AI公司集体重新包装
1715

1816
最近一堆AI公司纷纷改口，不再把"AGI"挂在嘴边了。The Verge敏锐地捕捉到了这个趋势——当人人都能宣称接近AGI的时候，AGI这个词就失去了意义。与其继续卷这个已经通货膨胀的概念，不如重新找一个能制造稀缺性的标签。
1917

2018
> 来源：The Verge
2119
22-
### 3. AI儿童玩具：一个正在失控的"狂野西部"
20+
## 3. AI儿童玩具：一个正在失控的"狂野西部"
2321

2422
Ars Technica报道了AI儿童玩具市场的乱象。没有统一标准、没有有效监管、没有明确的数据安全要求。给孩子买一个"AI玩偶"，你可能同时买回了隐私泄露和不可控的内容生成。家长们需要知道，这个领域目前几乎是裸奔状态。
2523

2624
> 来源：Ars Technica
2725
28-
### 4. Google要在AI搜索概览里放更多来源链接了
26+
## 4. Google要在AI搜索概览里放更多来源链接了
2927

3028
Google终于意识到，AI Overviews如果只给答案不给出处，正在把整个内容生态推向死路。Google宣布将在AI生成的搜索概览中加入更多网站链接。这不算彻底解决，但至少是向正确方向迈了一步——让流量回流到创作者手里，而不是全被摘要吃掉。
3129

‎src/content/en/daily/2026-05-09.md‎

Lines changed: 4 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -5,27 +5,25 @@ tags: ["Nvidia", "DeepSeek", "Anthropic", "OpenAI", "Baidu"]
55
summary: "Nvidia invests $40B in AI equity deals; DeepSeek raises $7B with V4.1 coming in June; Baidu launches ERNIE 5.1; Cloudflare admits AI made 1,100 jobs obsolete."
66
---
77

8-
## Top Stories
9-
10-
### 1. Nvidia Has Already Poured $40 Billion Into AI Equity Investments This Year
8+
## 1. Nvidia Has Already Poured $40 Billion Into AI Equity Investments This Year
119

1210
Nvidia's pace of investment in AI infrastructure is accelerating. According to TechCrunch, in just the first few months of this year, Nvidia has already committed $40 billion to AI equity deals. That's more aggressive than many expected — Jensen Huang isn't just selling chips, he's using capital to tie the entire AI ecosystem to his platform. From chips to ecosystem, the moat is getting deeper.
1311

1412
> Source: TechCrunch
1513
16-
### 2. DeepSeek Records $7B Funding, V4.1 Scheduled for June
14+
## 2. DeepSeek Records $7B Funding, V4.1 Scheduled for June
1715

1816
According to QbitAI, Liang Wenfeng personally contributed 20 billion yuan to lead DeepSeek's first round of 50 billion yuan ($7 billion) in funding, setting a record for Chinese AI companies. The V4.1 release is already scheduled for June. Last year, DeepSeek carved out its own path internationally with open-source models. Now with money and technology, the real question is how it'll go head-to-head with closed-source rivals.
1917

2018
> Source: QbitAI
2119
22-
### 3. Baidu Launches ERNIE 5.1: Tops Domestic Search, Training Cost Only 6% of Industry
20+
## 3. Baidu Launches ERNIE 5.1: Tops Domestic Search, Training Cost Only 6% of Industry
2321

2422
Baidu released ERNIE 5.1 today, with search capabilities ranking first among all domestic large models. What's even more noteworthy is the pre-training cost — Baidu claims it's only 6% of the industry standard. If true, this means Baidu has significantly optimized its underlying training efficiency. As the LLM arms race enters a phase where efficiency matters more than raw parameter count, cost control may be more persuasive than scale.
2523

2624
> Source: QbitAI
2725
28-
### 4. Cloudflare Admits AI Made 1,100 Jobs Obsolete, Even as Revenue Hits Record Highs
26+
## 4. Cloudflare Admits AI Made 1,100 Jobs Obsolete, Even as Revenue Hits Record Highs
2927

3028
Record revenue on one side, AI replacing 1,100 jobs on the other. Cloudflare's candid statement is representative of a broader reality — AI's impact on employment is no longer a prediction, it's happening now. Interestingly, the company didn't try to hide it, instead presenting it alongside its performance growth.
3129

‎src/content/en/daily/2026-05-10.md‎

Lines changed: 4 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -5,27 +5,25 @@ tags: ["Nvidia", "AGI", "AI Regulation", "OpenAI", "Google AI"]
55
summary: "Nvidia commits $40B to AI equity deals this year; AGI concept gets collectively rebranded; Google adds more source links to AI Overviews"
66
---
77

8-
## Top Stories
9-
10-
### 1. Nvidia has already dropped $40B on AI equity deals this year
8+
## 1. Nvidia has already dropped $40B on AI equity deals this year
119

1210
We're not even halfway through the year, and Nvidia has committed $40 billion in equity AI investments. This isn't just about selling chips anymore — Nvidia is buying its way into the entire AI supply chain. The more companies it invests in, the deeper the moat gets. At this point, Nvidia isn't just powering AI, it's financing it.
1311

1412
> Source: TechCrunch
1513
16-
### 2. It's the great AGI rebrand
14+
## 2. It's the great AGI rebrand
1715

1816
A bunch of AI companies are quietly dropping "AGI" from their messaging. The Verge caught the trend: when everyone claims to be close to AGI, the word means nothing. So now the industry is searching for the next scarcity-driven label. It's the classic move — rebrand before the current term completely inflates.
1917

2018
> Source: The Verge
2119
22-
### 3. AI kids' toys: the new Wild West
20+
## 3. AI kids' toys: the new Wild West
2321

2422
Ars Technica reports on the barely-regulated market of AI-powered children's toys. No standards, no meaningful oversight, and murky data privacy practices. Buying an "AI doll" for your kid might also mean buying a privacy leak and an unpredictable content generator. Parents should know: this space is essentially running naked right now.
2523

2624
> Source: Ars Technica
2725
28-
### 4. Google will add more source links to AI Overviews
26+
## 4. Google will add more source links to AI Overviews
2927

3028
Google is finally acknowledging a problem it helped create: AI-generated search summaries that give answers without giving credit. The fix? More links to original sources in AI Overviews. It's not a complete solution, but it's a step in the right direction — letting traffic flow back to creators instead of getting swallowed by AI summaries.
3129

0 commit comments

Comments
 (0)