高频日报

数据 社区

技能解读

当前原文的自动摘译,尚未人工审校。

何时使用

当用户要求:提取某天的Hugging Face每日论文列表 获取某天日报的标题、点赞量、arXiv ID、评论数、GitHub Star 等领域分类写作论文清单 将结果保存为 Markdown 文档时,请使用此技能

批判性的洞察力

HF Daily Papers 页面是一个 Svelte SPA。服务器呈现的 HTML 不包含人类可读的纸质数据。相反,所有纸质元数据都位于一个 JSON blob 中data-props属性:请勿使用web_fetch对于此页面。 Markdown/文本转换会丢失 arXiv ID、作者链接和结构化元数据。相反,使用download_file获取原始 HTML,然后使用 Python 提取 JSON(见下文)。

1)确定目标日期

如果未指定日期,则默认为昨天。页面使用格式:https://huggingface.co/papers/date/YYYY-MM-DD日报仅在工作日更新。如果用户请求周末约会,请警告该列表可能尚未更新。

2) 下载原始 HTML

使用download_file(不web_fetch):这会保留完整的 DOM,包括data-propsJSON 斑点。

3)用Python提取JSON

使用捆绑的脚本或以下内联 Python 模式:

流狐档案 运行与权限由流狐静态检查推断,不代表作者声明或实际测试结果。
领域
数据
作者声明的智能体
未找到明确声明;不据此推断已兼容或已测试
静态检查
88 / 100 · 启发式扫描,不代表运行安全
运行环境证据
未执行;当前没有隔离运行测试证据
作者 / 版本 / 许可证
@tomevault-io · 未声明许可证
上下文用量评级
低消耗
接入复杂程度
即装即用
是否需要外部接口密钥
不需要
兼容的系统
未声明(默认跨平台)
运行环境要求
Python
检测到的文件与系统行为
  • 只读
  • 允许写入 / 修改
网络行为
允许外网请求
安装命令数
无(仅作为资料)

需要注意: 未声明工具权限范围;实际权限由运行环境决定。

作者说明文件(原文) SKILL.md

# Hugging Face Daily Papers

Fetch Hugging Face Daily Papers for a target date, extract per-paper full metadata from the embedded Svelte hydration JSON, classify papers by domain, and save the result as a Markdown + JSON report.

## When to use

Use this skill when the user asks to:
- 抓取某天的 Hugging Face 每日论文列表
- 获取某天 Daily Papers 的标题、点赞量、arXiv ID、评论数、GitHub Star 等
- 按领域分类输出论文清单
- 将结果保存为 Markdown 文档

## Critical insight

The HF Daily Papers page is a **Svelte SPA**. The server-rendered HTML does **not** contain human-readable paper data. Instead, all paper metadata lives in a single JSON blob inside a `data-props` attribute:

```html
<div class="SVELTE_HYDRATER contents"
     data-target="DailyPapers"
     data-props='{"dailyPapers":[{...21 papers...}], ...}'>
```

**Do NOT use `web_fetch` for this page.** The Markdown/text conversion loses arXiv IDs, author links, and structured metadata. Instead, use `download_file` to get the raw HTML, then extract JSON with Python (see below).

## Workflow

### 1) Determine the target date

- Default to yesterday if no date specified.
- The page uses format: `https://huggingface.co/papers/date/YYYY-MM-DD`
- Daily Papers are updated on workdays only. If the user requests a weekend date, warn that the list may not have been updated.

### 2) Download raw HTML

Use `download_file` (NOT `web_fetch`):

```
download_file(
  url="https://huggingface.co/papers/date/YYYY-MM-DD",
  dest="hf_papers_YYYY-MM-DD.html",
  overwrite=true
)
```

This preserves the full DOM including the `data-props` JSON blob.

### 3) Extract JSON with Python

Use the bundled script or the following inline Python pattern:

```python
import json, re

with open("hf_papers_YYYY-MM-DD.html", "r", encoding="utf-8") as f:
    html = f.read()

# Locate the JSON blob inside data-props
dp_start = html.rfind('data-props="', 0, html.find('"dailyPapers"') + 50000)
json_start = dp_start + len('data-props="')
json_end = html.find('"><section', json_start)

json_str = html[json_start:json_end]
json_str = json_str.replace("&quot;", '"').replace("&amp;", "&")

data = json.loads(json_str)
papers = data["dailyPapers"]
```

### 4) Extract per-paper fields

From each entry in `data.dailyPapers[i]`:

| Field | Path | Example |
|-------|------|------|
| arXiv ID | `paper.id` | `"2605.00658"` |
| Title | `title` (or `paper.title`) | `"UniVidX: A Unified..."` |
| Upvotes | `paper.upvotes` | `71` |
| Comments | `numComments` | `2` |
| GitHub Stars | `paper.githubStars` | `52` (may be absent) |
| Organization | `organization.name` | `"ByteDance"` (may be absent) |
| HF URL | `https://huggingface.co/papers/{paper.id}` | — |
| arXiv URL | `https://arxiv.org/abs/{paper.id}` | — |
| AI Summary | `paper.ai_summary` | Short one-liner |
| Keywords | `paper.ai_keywords` | Array of strings |
| Authors | `paper.authors[].name` | Array |
| Author participation | `isAuthorParticipating` | `true`/`false` |
| Submitted by | `submittedBy.fullname` | `"taesiri"` |
| Thumbnail | `thumbnail` | Image URL |

### 5) Save output files

Save both a structured JSON and a polished Markdown report:

| File | Format | Purpose |
|------|--------|------|
| `hf_daily_YYYY-MM-DD.json` | JSON | Machine-readable, full metadata |
| `hf_daily_YYYY-MM-DD.md` | Markdown | Human-readable report, domain-classified |

### 6) Classify by domain

Use the paper's primary contribution to assign one main domain. Common domains include:
- Agent / AI Systems / Tool Use / RAG
- Reinforcement Learning / Reward Modeling
- LLM Training / Distillation / Inference Efficiency / Safety
- Vision-Language / Multimodal / Robotics
- 3D / Graphics / World Models / Video Generation
- Memory / Cognitive Architectures
- Continual / Incremental Learning
- Healthcare / Biology
- Benchmarks / Evaluation

**Each paper gets exactly one domain.** Total entries must equal the total paper count.

## Output structure (Markdown)

```md
# 🤗 Hugging Face Daily Papers — YYYY-MM-DD

**共 N 篇论文** | 数据来源: https://huggingface.co/papers/date/YYYY-MM-DD

| # | 👍 | arXiv ID | 论文标题 | 💬 | ⭐ | 机构 |
|:--:|:--:|:--|------|:--:|:--:|------|
| 1 | 71 | `2605.00658` | [UniVidX: ...](https://arxiv.org/abs/2605.00658) | 2 | 52 | — |

---

## 论文速览

### 1. Paper Title
**arXiv:** `2605.xxxxx` | **👍 71** | **🏢 Organization**

> AI summary one-liner

**关键词:** kw1 · kw2 · kw3
---
```

## Quality bar

- Do NOT invent arXiv IDs. Every ID must come from the JSON.
- Do NOT use `web_fetch` for this task — use `download_file` + Python.
- Verify the paper count: `len(papers)` must equal 21 (or the day's actual count).
- Keep each overview concise and factual.
- Preserve the original paper title verbatim.
- Cross-check: the total entries in the Markdown table must equal the paper count in the header.

## Update cadence

- Hugging Face Daily Papers are typically updated on workdays only.
- If the user requests a weekend date, first check whether a date page exists. If not, report that the list may not have been updated that day and ask whether to use the previous workday.

## Bundled resources

- Reference: `references/extraction-notes.md` — detailed extraction method and edge cases

Read `references/extraction-notes.md` for the full Python extraction script and troubleshooting guide.

---
> Source: [ChangweiXu/PyClaego](https://github.com/ChangweiXu/PyClaego) — distributed by [TomeVault](https://tomevault.io).
<!-- tomevault:4.0:skill_md:2026-05-22 -->

讨论

基于 GitHub Discussions。登录 GitHub 即可参与讨论、点赞、订阅更新。