Description

具身智能周报自动化生成与发布技能。当用户需要生成具身智能领域一周动态报告时使用，覆盖完整工作流：ArXiv论文多方向检索与整理、GitHub开源项目趋势追踪、综合可视化HTML报告生成（含导读+统计图表），以及将报告推送到 GitHub Pages 仓库发布。适用于每周定期生成具身智能领域动态报告的场景。

README (SKILL.md)

\r \r

Embodied AI Weekly 具身智能周报生成 Skill\r

Name: Embodied Ai Weekly
Author: jessy-huang

\r

概述\r

\r 本 Skill 封装了具身智能周报从「信息检索」到「报告生成」再到「GitHub 发布」的完整 SOP，分三个阶段：\r \r

Phase 1 — ArXiv 论文检索：按 7 个研究方向检索最新论文\r
Phase 2 — GitHub 开源项目检索：按 7 个方向追踪新增仓库与经典仓库动态\r
Phase 3 — 综合报告生成与发布：整合两部分内容，生成 Markdown + HTML 综合报告并推送至 GitHub\r \r ---\r \r

Phase 1 — ArXiv 论文检索\r

\r

检索策略\r

\r 使用 web_fetch 工具逐一检索 7 个方向，每个方向独立请求。参考 references/arxiv_search_guide.md 获取各方向的检索关键词与 API URL 模板。\r \r API URL 模板：\r

https://arxiv.org/search/?query=\x3CKEYWORDS>&searchtype=all&start=0&order=-announced_date_first\r
```\r
\r
或使用 arXiv 官方 RSS / cs.RO 分类页面：\r
```\r
https://arxiv.org/list/cs.RO/recent\r
https://arxiv.org/list/cs.CV/recent\r
```\r
\r
### 检索方向（共 7 个）\r
\r
每个方向的详细关键词见 `references/arxiv_search_guide.md`：\r
\r
| 序号 | 方向 | 核心关键词（示例） |\r
|------|------|-------------------|\r
| 1 | 具身感知与场景理解 | egocentric perception, affordance, 3D scene graph |\r
| 2 | 具身决策与规划 | embodied planning, LLM robot, TAMP, long-horizon |\r
| 3 | 具身控制与操作 | dexterous manipulation, diffusion policy, visuomotor |\r
| 4 | 具身强化学习与世界模型 | embodied RL, world model robot, sim-to-real |\r
| 5 | 具身智能体与大模型 | VLA, vision-language-action, OpenVLA, LLM robotics |\r
| 6 | 仿真、数据与平台 | robotic simulation, embodied benchmark, x-embodiment |\r
| 7 | 人机交互与具身社会智能 | human-robot interaction, shared autonomy, HRI |\r
\r
### 输出格式\r
\r
为每篇论文整理以下字段：\r
- 标题（中英对照）\r
- ArXiv 链接\r
- 作者与机构\r
- 核心贡献（200字以内）\r
- 方向标签\r
\r
生成文件：`embodied_ai_weekly_arxiv_\x3CYEAR>_w\x3CWEEK>.md` 和对应 `.html`\r
\r
---\r
\r
## Phase 2 — GitHub 开源项目检索\r
\r
### 检索策略\r
\r
使用 `web_fetch` 工具，检索两类仓库：\r
- **新增仓库**：`created:>YYYY-MM-DD`（过去7天），按 stars 排序\r
- **经典仓库**：`pushed:>YYYY-MM-DD`，关注有重要更新的高星仓库\r
\r
**GitHub API URL 模板：**\r
```\r
https://api.github.com/search/repositories?q=\x3CKEYWORDS>&sort=stars&order=desc&per_page=10\r
```\r
\r
> ⚠️ 注意：GitHub API 对 OR 运算符有限制（最多 5 个），需要拆分查询。若 API 超时，改用 GitHub 网页搜索：`https://github.com/search?q=\x3CKEYWORDS>&type=repositories&s=updated&order=desc`\r
\r
### 检索方向\r
\r
与 ArXiv 对应的 7 个方向，详见 `references/github_search_guide.md`\r
\r
### 输出格式\r
\r
每个仓库整理：\r
- 仓库名 + GitHub 链接\r
- 简介（50字内）\r
- Stars 数量\r
- 是否为本周新增 / 经典更新\r
- 所属方向\r
\r
生成文件：`embodied_ai_weekly_github_\x3CYEAR>_w\x3CWEEK>.md` 和对应 `.html`\r
\r
---\r
\r
## Phase 3 — 综合报告生成与发布\r
\r
### 3.1 综合 Markdown 报告\r
\r
读取 Phase 1 和 Phase 2 的 `.md` 文件，整合为一份综合报告，结构如下：\r
\r
```\r
# 具身智能深度周报 \x3CYEAR>-W\x3CWEEK>\r
> 报告周期：\x3CDATE_START> ~ \x3CDATE_END>\r
\r
## 📖 一周导读\r
- 快速导航（7方向跳转链接）\r
- 本周 Top 5 必读（论文 + 项目）\r
- 核心趋势卡片（5大趋势）\r
- 开放问题（4个待关注点）\r
\r
## 📊 数据总览\r
- 论文方向分布表格\r
- GitHub 仓库统计表格\r
- 每日发表趋势\r
\r
## 📄 ArXiv 论文精选\r
（7方向，每方向列出所有检索论文，精选论文配 ArXiv 原文 figure）\r
\r
## 🐙 GitHub 开源项目\r
（新建仓库精选 + 经典仓库重点更新，配项目截图）\r
\r
## 💡 核心洞察\r
（5大趋势分析 + 下周关注点）\r
```\r
\r
生成文件：`embodied_ai_weekly_report_\x3CYEAR>_w\x3CWEEK>.md`\r
\r
### 3.2 论文配图获取\r
\r
报告需要配图来增强可读性。配图来源为 ArXiv 论文原文和 GitHub 项目页面，**禁止使用 AI 生成图片**。\r
\r
**图片获取策略：**\r
\r
1. **ArXiv 论文图片**：使用 Playwright 或直接 HTTP 请求访问论文 HTML 版本，下载其中的 figure 图片\r
   - HTML 版本 URL 模板：`https://arxiv.org/html/\x3CARXIV_ID>v\x3CVERSION>/`\r
   - 图片通常为 `x1.png`、`x2.png` 等命名\r
   - 可直接用 HTTP 下载：`https://arxiv.org/html/\x3CARXIV_ID>v\x3CVERSION>/x1.png`\r
   - 保存到工作区 `images/` 目录\r
\r
2. **GitHub 项目截图**：使用 Playwright 对 GitHub 仓库页面进行截图\r
   - 截取 README Banner 或项目首页区域\r
   - 保存到 `images/` 目录\r
\r
3. **论文项目页截图**：部分论文有配套项目页（如 `https://faceong.github.io/CoEnv/`），可直接截图\r
\r
**图片嵌入位置：**\r
- **封面区（Hero）**：选择当周最重要/最热门的论文项目页截图\r
- **精选论文卡片**：每篇精选论文配 1-2 张 ArXiv 原文 figure\r
- **GitHub 热榜卡片**：配 GitHub 仓库截图\r
- 每张图片附带 `\x3Cfigcaption>` 注明来源链接\r
\r
**批量下载脚本示例（Node.js + Playwright）：**\r
\r
```javascript\r
const { chromium } = require('playwright');\r
// 使用 Playwright 截图 GitHub 页面\r
// 使用 fetch/axios 下载 ArXiv 论文图片\r
// 保存到 images/ 目录\r
```\r
\r
**图片 CSS 样式（添加到 HTML 内联样式）：**\r
\r
```css\r
.paper-figure { margin: 14px 0; border-radius: 10px; overflow: hidden; border: 1px solid var(--border); }\r
.paper-figure img { width: 100%; display: block; max-height: 400px; object-fit: contain; background: #fff; }\r
.paper-figure figcaption { padding: 8px 14px; font-size: 11px; color: var(--text3); text-align: center; }\r
.hero-figure { margin: 30px auto; max-width: 900px; border-radius: 16px; overflow: hidden; box-shadow: 0 8px 40px rgba(99,102,241,.15); }\r
.hero-figure img { width: 100%; display: block; }\r
.github-figure { margin: 12px 0; border-radius: 8px; overflow: hidden; border: 1px solid var(--border); }\r
.github-figure img { width: 100%; display: block; max-height: 320px; object-fit: cover; }\r
```\r
\r
### 3.3 综合 HTML 可视化报告\r
\r
HTML 报告的视觉风格参考 `references/html_template_guide.md`。\r
\r
**必须包含的模块：**\r
\r
1. **Header Hero 区**：渐变标题 + 报告日期 + Executive Summary + 封面图\r
2. **Stats Cards 数据卡片**：论文数、GitHub 仓库数、方向数等\r
3. **ArXiv 分布条形图**：使用 Chart.js 或纯 CSS 条形图展示7方向论文分布\r
4. **论文卡片区**：Top 3-5 精选论文深度解读 + 每篇配图 + 完整论文表格（可折叠）\r
5. **GitHub 热榜**：Top 3 仓库卡片 + 仓库截图 + 全方向仓库矩阵\r
6. **核心洞察区**：5大趋势 + 下周关注\r
\r
**技术规范：**\r
- 自包含（不依赖外部 CSS 框架，内联所有样式）\r
- 暗色主题（背景 `#0B0E14`，强调色 indigo/cyan/neon）\r
- 使用 Chart.js CDN：`https://cdn.jsdelivr.net/npm/chart.js`\r
- 支持滚动动画（IntersectionObserver）\r
- 响应式布局（移动端适配）\r
\r
生成文件：`embodied_ai_weekly_report_\x3CYEAR>_w\x3CWEEK>.html`\r
\r
### 3.4 推送到 GitHub Pages\r
\r
**目标仓库结构：**\r
```\r
\x3Cgithub-repo>/\r
├── \x3CYEAR>-W\x3CWEEK>/\r
│   ├── index.html     ← 综合 HTML 报告\r
│   └── report.md      ← Markdown 全文\r
├── archive/\r
│   └── index.html     ← 归档索引（需更新）\r
└── latest/\r
    └── index.html     ← 最新周报入口（需更新为重定向）\r
```\r
\r
**推送流程：**\r
\r
```bash\r
# 1. 克隆仓库（优先 SSH，避免 HTTPS 网络问题）\r
git clone [email protected]:\x3CUSER>/\x3CREPO>.git \x3CLOCAL_PATH> --depth 1\r
\r
# 2. 创建当周文件夹\r
mkdir \x3CLOCAL_PATH>/\x3CYEAR>-W\x3CWEEK>\r
\r
# 3. 复制文件\r
cp \x3CREPORT>.html \x3CLOCAL_PATH>/\x3CYEAR>-W\x3CWEEK>/index.html\r
cp \x3CREPORT>.md   \x3CLOCAL_PATH>/\x3CYEAR>-W\x3CWEEK>/report.md\r
\r
# 4. 更新 latest/index.html（改为 meta refresh 重定向到新周）\r
# 5. 更新 archive/index.html（在列表顶部插入新周条目）\r
\r
# 6. 提交推送\r
git add -A\r
git commit -m "feat: add W\x3CWEEK> report (\x3CDATE_START> ~ \x3CDATE_END>)"\r
git push origin main\r
```\r
\r
> ⚠️ HTTPS 推送若超时，改用 SSH (`[email protected]:...`)\r
\r
**latest/index.html 模板（重定向）：**\r
```html\r
\x3C!DOCTYPE html>\r
\x3Chtml lang="zh-CN">\r
\x3Chead>\r
  \x3Cmeta charset="UTF-8">\r
  \x3Cmeta http-equiv="refresh" content="0; url=./\x3CYEAR>-W\x3CWEEK>/index.html">\r
  \x3Ctitle>重定向...\x3C/title>\r
\x3C/head>\r
\x3Cbody>\x3Cp>\x3Ca href="./\x3CYEAR>-W\x3CWEEK>/index.html">点击跳转\x3C/a>\x3C/p>\x3C/body>\r
\x3C/html>\r
```\r
\r
**archive/index.html 新增条目模板：**\r
```html\r
\x3Cdiv class="archive-item">\r
  \x3Cspan class="archive-date">📅 \x3CYEAR>-W\x3CWEEK> · \x3CDATE_START> ~ \x3CDATE_END>\x3C/span>\r
  \x3Ca href="../\x3CYEAR>-W\x3CWEEK>/index.html" class="archive-link">查看详情 →\x3C/a>\r
\x3C/div>\r
```\r
\r
---\r
\r
## 输出文件清单\r
\r
| 文件 | 内容 | 位置 |\r
|------|------|------|\r
| `images/` | 论文配图与项目截图目录 | 工作区根目录 |\r
| `embodied_ai_weekly_arxiv_\x3CY>_w\x3CW>.md` | ArXiv 论文整理 | 工作区根目录 |\r
| `embodied_ai_weekly_arxiv_\x3CY>_w\x3CW>.html` | ArXiv 报告网页版 | 工作区根目录 |\r
| `embodied_ai_weekly_github_\x3CY>_w\x3CW>.md` | GitHub 项目整理 | 工作区根目录 |\r
| `embodied_ai_weekly_github_\x3CY>_w\x3CW>.html` | GitHub 报告网页版 | 工作区根目录 |\r
| `embodied_ai_weekly_report_\x3CY>_w\x3CW>.md` | 综合 Markdown 报告 | 工作区根目录 |\r
| `embodied_ai_weekly_report_\x3CY>_w\x3CW>.html` | 综合可视化 HTML 报告 | 工作区根目录 |\r
\r
---\r
\r
## 用户配置信息\r
\r
在开始前，需确认以下信息（若未提供则询问用户）：\r
\r
| 参数 | 说明 | 示例 |\r
|------|------|------|\r
| 报告周期 | 起止日期 | 2026-03-26 ~ 2026-04-02 |\r
| 周编号 | ISO 周数 | W13 |\r
| GitHub 仓库 | 发布目标仓库 | `Jessy-Huang/embodied-ai-weekly` |\r
| 推送方式 | SSH 或 HTTPS | SSH（推荐） |\r
| 工作区路径 | 报告文件保存路径 | 当前 WorkBuddy 工作区 |\r

Usage Guidance

Before installing or running this skill, consider the following: - Tooling and capabilities required: The skill expects to run web requests, download images, run Playwright (headless browser) and Node.js scripts, and use git to clone/commit/push. Confirm these tools are available in the execution environment — the skill metadata does not declare them. - GitHub authentication: Publishing requires GitHub credentials (SSH key or HTTPS token). The skill does not request or document how credentials are supplied or stored. Only grant repository push rights if you trust the exact behavior; prefer using a dedicated repo or deploy key with limited permissions rather than your main account token. - Review push behavior: The instructions assume the agent will update latest/index.html and archive/index.html and then push to main. Decide whether you want the agent to push automatically or require manual review/PR. If you want safer behavior, modify the workflow to create a branch and open a PR instead of pushing to main. - Data & copyright: The skill downloads figures from ArXiv and screenshots GitHub/project pages. Ensure you have the right to redistribute images and that you comply with arXiv/github content policies and any paper authors' terms. - Rate limits and tokens: GitHub API calls may require a GITHUB_TOKEN to avoid strict rate limits; arXiv scraping may also be throttled. Confirm how the agent will handle failures and whether credentials will be used. - If you plan to run this in a shared or hosted agent environment, confirm sandboxing and that Playwright screenshots will not capture sensitive pages. Limit the agent's network access and filesystem write scope where possible. What would change this assessment: explicit metadata declaring required binaries (git, node, playwright), required env vars (e.g., GITHUB_TOKEN or instructions about SSH deploy keys), and a safer publish flow (branch + PR) or confirmation prompts before pushing would make the skill coherent and reduce the concerns.

Capability Analysis

Type: OpenClaw Skill Name: embodied-ai-weekly Version: 1.1.0 The skill automates a complex workflow involving web scraping (ArXiv/GitHub), automated browser control via Playwright for screenshots, and high-privilege Git operations including 'git push' to a user-specified repository. While these capabilities are aligned with the stated purpose of generating and publishing a weekly report to GitHub Pages, the requirement for Git credentials and the ability to modify remote repositories represent a significant risk. No evidence of intentional malice, data exfiltration, or hidden prompt injections was found in SKILL.md or the reference guides, but the broad permissions required for Phase 3 (publication) warrant a suspicious classification.

Capability Assessment

⚠ Purpose & Capability

The skill claims to generate and publish weekly reports (ArXiv + GitHub → HTML → GitHub Pages), which is coherent with the instructions. However, the instructions assume availability of tools (Node.js/Playwright, git, ability to write a workspace and perform network requests and screenshots) but the skill declares no required binaries, env vars, or credentials. A publishing workflow that clones and pushes to a GitHub repo generally requires Git and credentials (SSH key or a token); those are not requested or documented as required. This mismatch between claimed capabilities and declared requirements is concerning.

⚠ Instruction Scope

SKILL.md explicitly directs the agent to: (a) fetch web pages and GitHub API endpoints (web_fetch), (b) download figures from ArXiv, (c) run Playwright to screenshot GitHub and project pages, (d) write files (images/, .md/.html), and (e) clone, commit, and push to a GitHub repository. All of those actions are within the stated purpose, but they expand the agent's operational scope beyond a pure read-only scraper: the skill will create files and push changes to a remote repository. The instructions do not document how credentials are provided, how rate limiting or API auth will be handled, or whether the agent will prompt before pushing — granting push rights implicitly is a sensitive action and should be made explicit.

ℹ Install Mechanism

There is no install spec (instruction-only), which lowers installer risk since nothing is automatically downloaded. However, SKILL.md examples reference Node.js and Playwright for screenshots and provide a Node.js snippet; the runtime therefore effectively depends on those packages being available. That dependency is not declared in the registry metadata. The lack of an install step is safe from an arbitrary-download perspective, but the skill will fail or require manual installation of tools; this mismatch is a notable usability and security gap.

⚠ Credentials

The skill declares no required environment variables or primary credential, yet its push workflow depends on GitHub authentication (SSH key or HTTPS token) and the GitHub API usage benefits from a GITHUB_TOKEN for rate limits. The absence of any declared credential requirement is disproportionate to the described behavior. Users must provide Git credentials (or the agent will attempt SSH), but the skill does not instruct how those credentials will be used, whether they will be stored, or whether the agent will request them interactively.

✓ Persistence & Privilege

The skill is not set to always:true and does not request persistent installation or elevated platform-wide privileges. Autonomous model invocation is not disabled (default), which is normal. There is no evidence the skill will modify other skills or system-wide settings.

Version History

v1.1.0

v1.1: 新增论文配图流程 - 从ArXiv原文和GitHub项目页获取真实配图并嵌入报告，优化HTML报告样式支持图片展示

v1.0.0

初始版本：四阶段完整工作流（ArXiv检索→GitHub追踪→综合报告→GitHub Pages发布）

Metadata

Slug embodied-ai-weekly

Version 1.1.0

License MIT-0

All-time Installs 0

Active Installs 0

Total Versions 2

Frequently Asked Questions

What is Embodied Ai Weekly?

具身智能周报自动化生成与发布技能。当用户需要生成具身智能领域一周动态报告时使用，覆盖完整工作流：ArXiv论文多方向检索与整理、GitHub开源项目趋势追踪、综合可视化HTML报告生成（含导读+统计图表），以及将报告推送到 GitHub Pages 仓库发布。适用于每周定期生成具身智能领域动态报告的场景。 It is an AI Agent Skill for Claude Code / OpenClaw, with 109 downloads so far.

How do I install Embodied Ai Weekly?

Run "/install embodied-ai-weekly" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Embodied Ai Weekly free?

Yes, Embodied Ai Weekly is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Embodied Ai Weekly support?

Embodied Ai Weekly is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Embodied Ai Weekly?

It is built and maintained by Jessy-Huang (@jessy-huang); the current version is v1.1.0.

More Skills

Embodied Ai Weekly