← Back to Skills Marketplace
mesevenjourney

Karpathy Curated RSS Brief

by ME柒的旅途 · GitHub ↗ · v1.0.2 · MIT-0
cross-platform ✓ Security Clean
504
Downloads
0
Stars
3
Active Installs
3
Versions
Install in OpenClaw
/install karpathy-curated-rss-brief
Description
Fetch articles from Karpathy's curated 93 RSS feeds and generate a Chinese tech daily newsletter. Triggers: RSS 日报、RSS 简报、karpathy-curated-rss-brief、每日简报、Kar...
README (SKILL.md)

Karpathy RSS 日报生成工作流

从 Andrej Karpathy 精选的 93 个 RSS 源中抓取最近文章,挑选最有价值的内容,生成一份高质量中文日报。

技能文件结构

\x3Cskill_dir>/
├── SKILL.md                       ← 本文件
├── scripts/
│   └── fetch_feeds.py             ← RSS 并发抓取脚本
└── references/
    └── output-template.md         ← 输出格式模板

订阅源 OPML 托管于 GitHub Pages,脚本运行时自动获取: https://mesevenjourney.github.io/static/hn-popular-blogs-2025.opml


工作流程

Step 1: 定位技能目录,运行抓取脚本

从上下文直接推导 SKILL_DIR,禁止使用 Glob/Grep 搜索。

本 SKILL.md 的路径会出现在对话上下文中(ide_opened_file 标签、Skill 系统加载记录、或 Read 工具的调用路径)。找到包含 karpathy-curated-rss-brief/SKILL.md 的那条记录,去掉末尾的 /SKILL.md,即为 SKILL_DIR

示例:若上下文中路径为 /path/to/skills/karpathy-curated-rss-brief/SKILL.md,则 SKILL_DIR=/path/to/skills/karpathy-curated-rss-brief

后续所有文件均基于此路径推导,无需再次搜索:

文件 路径
抓取脚本 SKILL_DIR/scripts/fetch_feeds.py
输出模板 SKILL_DIR/references/output-template.md
订阅源 SKILL_DIR/hn-popular-blogs-2025.opml(脚本内部自动读取)

运行抓取脚本:

uv run --script \x3CSKILL_DIR>/scripts/fetch_feeds.py --hours 24

如果 24 小时内文章不足 5 篇,改用 --hours 48。脚本自动读取内置 OPML,输出最多 20 篇文章(按发布时间倒序)到 stdout,日志到 stderr:

[{"title": "...", "link": "...", "author": "...", "source": "...", "published": "...", "summary": "..."}]

脚本使用 PEP 723 inline metadata,uv run --script 自动安装依赖,无需手动 pip install。

Step 2: 筛选最有价值的文章

从抓取结果中挑选 8-10 篇。筛选标准:

🔴 最高优先级(只要出现,必须纳入):

  • AI 重大新闻:LLM 新模型发布(GPT、Claude、Gemini、Llama、Qwen 等)、重大模型能力突破、主流 AI 工具/框架的重要版本更新
  • 重磅硬件/产品发布:Apple 新品(iPhone、Mac、iPad、Vision Pro 等)、Google 硬件/平台、Windows PC 新品、Nvidia GPU/芯片发布、其他影响消费级或企业级市场的重大产品发布

🟡 次优先级(酌情选入):

  1. 话题多样性 — 覆盖不同领域(AI、安全、编程、科技政策等),避免同一话题重复
  2. 深度优先 — 有独特观点和深度分析,而非简短资讯
  3. 时效性 — 最新发布的优先
  4. 影响力 — 知名作者或重要话题优先

Step 3: 获取文章全文

对每篇选中的文章,使用 WebFetch 工具读取全文。编者观察必须基于全文,不能仅依赖 RSS 摘要。

Step 4: 按模板生成中文日报

Read 工具读取模板文件:SKILL_DIR/references/output-template.md,严格按其格式生成日报。

分类 emoji 参考:🤖 AI/ML、⚠️ 安全/风险、🧠 认知/思维、📚 技术/编程、📜 政策/法规、🔧 工具/开源、🌐 互联网/平台、💰 商业/创业

Step 5: 撰写编者观察(核心环节)

💡 编者观察需要真正的综合思考,基于已完整阅读的所有文章,从以下维度思考后再下笔:

  1. 横向联系:不同领域的文章有哪些隐藏关联?表面无关的事件是否指向同一深层趋势?
  2. 时代信号:这批内容传递了什么尚未被主流关注的信号?
  3. 反直觉洞察:综合所有内容后,有什么读者单独读每篇都不会注意到的结论?
  4. 值得追踪:哪些苗头现在看起来小,但可能在 6-12 个月内变成大趋势?

写法要求:

  • 1-2 段,每段 3-5 句,观点密度高
  • 禁止"今日文章涵盖 X、Y、Z 主题"这类罗列句式
  • 有明确的、可辩驳的观点,读者看完应有"没想到"的感受
  • 用第一人称("我注意到..."),体现真正的编辑判断

Step 6: 保存文件并校验编码

将日报保存到当前工作目录,文件名:

{YYYY-MM-DD}-Karpathy精选RSS日报.md

保存后立即运行编码校验,检测 Unicode 替换字符(U+FFFD)——这是 LLM 生成长中文文本时偶发的已知问题:

python3 -c "
import sys
path = sys.argv[1]
text = open(path, encoding='utf-8').read()
bad = [(i+1, line) for i, line in enumerate(text.splitlines()) if '\ufffd' in line]
if bad:
    print(f'❌ 发现 {len(bad)} 处乱码:')
    for lineno, line in bad:
        print(f'  第{lineno}行: {line}')
    sys.exit(1)
else:
    print('✅ 编码校验通过,无乱码')
" {输出文件路径}

如果校验失败:逐行对照原文重新填写被替换的汉字,再次运行校验直到通过,不要跳过此步骤

Usage Guidance
This skill appears to do exactly what it says: fetch a curated OPML, retrieve recent feed entries, read article pages, and produce a Markdown newsletter. Before installing, consider: 1) The OPML is fetched at runtime from mesevenjourney.github.io — review that OPML (or host a local copy) to ensure the feed list is what you expect, since the list determines which external sites the skill will visit. 2) 'uv run --script' will install Python packages (feedparser, aiohttp) at run time — if you restrict package installs, run the script in a controlled environment. 3) The skill writes files to the agent's current working directory; ensure that path is acceptable. 4) The WebFetch step will retrieve full article pages from arbitrary external sites listed in the OPML — be aware of network traffic and privacy implications. If you want stronger guarantees, inspect the OPML and the included fetch_feeds.py (already present) and, if desired, replace the remote OPML with a vetted local copy.
Capability Analysis
Type: OpenClaw Skill Name: karpathy-curated-rss-brief Version: 1.0.2 The skill is designed to fetch and summarize RSS feeds from a curated list attributed to Andrej Karpathy. It uses a Python script (fetch_feeds.py) to asynchronously retrieve articles and provides clear instructions in SKILL.md for the agent to filter, fetch full text, and format a newsletter. No evidence of data exfiltration, unauthorized execution, or malicious prompt injection was found; the included Python validation snippet is a functional utility to handle common LLM encoding issues.
Capability Assessment
Purpose & Capability
Name/description match the included code and instructions: a Python fetch script plus an output template and guidance to generate a newsletter. No credentials, binaries, or unrelated dependencies are requested. Minor note: SKILL.md lists a local OPML path (SKILL_DIR/hn-popular-blogs-2025.opml) but the actual script fetches the OPML from a GitHub Pages URL — functionally equivalent but slightly inconsistent in wording.
Instruction Scope
SKILL.md confines actions to fetching the OPML/feed list, running the included fetch_feeds.py, using a WebFetch tool to retrieve article pages, producing a Markdown file, and validating encoding. It does not instruct reading other system files, environment variables, or sending data to third-party endpoints other than the OPML host and discovered feed URLs. One stylistic oddity: it requires deriving SKILL_DIR from conversation context rather than filesystem search, but this is an operational preference, not a secret-reading directive.
Install Mechanism
No separate install spec; the script relies on PEP 723 inline metadata and the platform's 'uv run --script' to install dependencies (feedparser, aiohttp). This is proportionate to the task (HTTP fetch + feed parsing). As with any runtime install, the platform will download packages from package indexes — expected for this functionality.
Credentials
The skill declares no required environment variables, no credentials, and no config paths. The script and SKILL.md do not attempt to read any other env vars or secrets. Network access is necessary and limited to the OPML URL and discovered feed/article URLs.
Persistence & Privilege
always is false and the skill is user-invocable. It does not request persistent agent-wide privileges or attempt to modify other skills. It writes output to the current working directory as part of its normal behavior (documented).
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install karpathy-curated-rss-brief
  3. After installation, invoke the skill by name or use /karpathy-curated-rss-brief
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.2
fix npx install cmd.
v1.0.1
Fetches opml from GitHub Pages at runtime.
v1.0.0
Karpathy精选RSS简报技能首次发布。 - 支持从Karpathy精选五93个RSS源抓取科技/AI相关最新文章。 - 自动筛选高价值资讯,生成每日8-10篇中文技术日报。 - 提供编者深度观察段落,强调多维趋势与非表面洞察。 - 输出严格符合模板规范,涵盖分类emoji、摘要与推荐理由。 - 内置编码校验,确保日报无乱码问题。
Metadata
Slug karpathy-curated-rss-brief
Version 1.0.2
License MIT-0
All-time Installs 3
Active Installs 3
Total Versions 3
Frequently Asked Questions

What is Karpathy Curated RSS Brief?

Fetch articles from Karpathy's curated 93 RSS feeds and generate a Chinese tech daily newsletter. Triggers: RSS 日报、RSS 简报、karpathy-curated-rss-brief、每日简报、Kar... It is an AI Agent Skill for Claude Code / OpenClaw, with 504 downloads so far.

How do I install Karpathy Curated RSS Brief?

Run "/install karpathy-curated-rss-brief" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Karpathy Curated RSS Brief free?

Yes, Karpathy Curated RSS Brief is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Karpathy Curated RSS Brief support?

Karpathy Curated RSS Brief is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Karpathy Curated RSS Brief?

It is built and maintained by ME柒的旅途 (@mesevenjourney); the current version is v1.0.2.

💬 Comments