ai-info-radar

name	ai-info-radar
description	AI news and resource discovery with subscription push. Triggers: AI media recommendations, AI podcasts, AI books, AI news, AI video recommendations, set up push notifications, subscribe to AI news, AI daily digest, scheduled AI push, Weibo AI influencers, latest from Jiqizhixin. Not for: real-time search; non-AI media.

ai-info-radar 8.0.0

First Use

Run the dependency check script before first use:

bash scripts/setup.sh

The agent will auto-run this on first trigger; usually no manual action needed.

Your all-in-one AI news assistant — aggregates 50+ AI media, blogs, podcasts, and book resources. Supports latest content fetching, smart recommendations, and scheduled push notifications.

---

Feature Overview

Feature	Description	Trigger Words
Resource Database	Query all AI learning resources	AI media, AI podcasts, AI books, AI videos
Latest Content	Fetch newest articles from sources	What did Jiqizhixin post recently, Search latest on OpenAI
Smart Recommendations	Recommend based on user background	AI resources for beginners, Technical AI resources
Bookmark Management	Track read/favorite items	Bookmark this, My AI reading list, My bookmarks
Scheduled Push	Custom frequency + topic notifications	Set up hourly Agent news push

When not to use

• Do not use for non-AI topics or general web search.
• Do not use when the user requires real-time completeness; use live web search and state the freshness window instead.
• Do not request cookies, tokens, login sessions, QR codes, or private account access for WeChat, X/Twitter, podcasts, or media sites.
• Do not present aggregated or third-party results as official source content.

Output contract

Match the user intent and always include:

• source category or platform;
• freshness window or last-check time when available;
• direct source links or a clear fallback/source caveat;
• recommendation reason, if recommending resources;
• push schedule, topic, and delivery target, if configuring subscriptions.

---

Trigger Words

Resource Queries

• AI媒体推荐, 推荐AI公众号, AI播客有哪些
• AI书籍推荐, 学AI看什么书
• AI视频推荐, AI课程推荐

Latest Content

• {media_name}最近发了什么 — e.g., What did Jiqizhixin post recently
• 搜 {keyword} 最新 — e.g., Search latest on Agent
• {media_name} 最新文章

Smart Recommendations

• 适合小白的AI资源, AI入门推荐
• 技术向AI媒体, 产品向AI资讯
• 英文AI资源, 中文AI资源

Bookmark Management

• 收藏这个, 加入收藏
• 我的收藏, 我的AI书单
• 取消收藏 {resource_name}

Scheduled Push

• 设置 {frequency} 推送 {topic} — e.g., Set up hourly Agent news push
• 设置推送, 订阅AI资讯, 每天推送AI资讯, 设置AI日报
• 查看我的推送设置
• 取消推送 {topic}
• 修改推送频率为 {frequency}

---

Supported Resources

Chinese Media / WeChat Accounts

Jiqizhixin (机器之心), Liangziwei (量子位), InfoQ AI, CSDN AI, AI前线, 新智元, AI科技评论, DataFunTalk, AI大模型实验室, 赛博禅心, BAAI (智源研究院), Mila Forum, AI研习社

Twitter/X

Fei-Fei Li, Andrej Karpathy, Yann LeCun, Mu Li, Lin Junyang, Dai Yusen, Baoyu, Guicang, swyx, Greg Brockman, Sam Altman, Demis Hassabis

Blogs / Websites

LessWrong, Import AI, The Batch, Papers With Code, Hugging Face Blog, Google AI Blog, OpenAI Blog, Anthropic Blog, Gradient Flow, Weights & Biases

Podcasts

42章经, 硅谷101, AI局内人, Latent Space, The TWIML AI Podcast, Lex Fridman Podcast

Books

Deep Learning (Goodfellow), Dive into Deep Learning (Li Mu), AIMA (Russell & Norvig), Life 3.0 (Tegmark), AI Superpowers (Kai-Fu Lee), Human Compatible (Russell)

Videos / Courses

Hung-Yi Lee ML, Andrew Ng Deep Learning, Mu Li Dive into DL, Andrej Karpathy Neural Networks, 3Blue1Brown, Two Minute Papers

---

Execution Flow

Step 1: Determine User Intent

• Query resource list → Return matching category content directly (see references/resource-list.md for full catalog)
• Fetch latest content → Enter Step 2
• Smart recommendation → Ask user background then recommend
• Bookmark management → Read/write ~/.ai-info-radar/favorites.json
• Set up push / Subscribe to AI news / Set up daily digest → Enter "Subscription + Push Flow (V3)"

Step 2: Fetch Latest Content

Select tools by platform type:

Platform	Primary Tool	Fallback Tool
WeChat Accounts	agent-reach fetch / Camoufox + HTTP proxy	Web search engines
Twitter/X	scripts/content_fetcher.py via xreach (agent-reach) → built-in guest token GraphQL fallback	Web search fallback
Weibo	Web search to locate UID/post URL + Playwright (visitor cookie)	—
Blogs/Websites	xread (agent-reach) → web_fetch / r.jina.ai	web_search
Bilibili/YouTube	yt-dlp / platform API	web_search
Content Search	mcporter/Exa (agent-reach) → Jina Search API (free)	DuckDuckGo HTML
Podcasts	iTunes Search API → RSS → download XML → parse latest episode	faster-whisper local transcription

📌 Xiaoyuzhou Podcasts: Use iTunes Search API to get RSS → download XML → parse latest episode. See references/podcast-xiaoyuzhou.md.

⚠️ WeChat Official Accounts may require HTTP proxy (HTTP_PROXY env var) in environments where direct access is blocked.

📌 Weibo: Chinese AI influencers (e.g., Jia Yangqing, Andrew Ng Chinese account) tracked via web search to locate UID/post URLs, then Playwright with visitor cookie for content. Text selector: .detail_wbtext_4CRf9. See references/weibo-source.md.

Twitter/X Fetch Example:

# Check AI researchers' latest tweets (top 10)
python3 - <<'PY'
from scripts.content_fetcher import ContentFetcher
fetcher = ContentFetcher()
for item in fetcher.fetch_from_twitter("karpathy", limit=10):
    print(item["published"], item["summary"][:180], item["url"])
PY

For full X profile lookup, pinned tweets, Note Tweet long-form handling, or JSON export, route to the dedicated x-twitter-scraper skill instead of assuming this repository has scripts/x_scraper.py.

Step 3: Content Processing & Output

After fetching, automatically:

• Deduplicate (same article across platforms, by URL + title similarity)
• Generate summaries (long articles compressed to 3-5 key points)
• Card format output (see format below)

---

Subscription + Push Flow (V3)

First-Time Push Setup

When user says "set up push", "subscribe to AI news", "daily AI digest":

Step 1: Ask for Configuration

• Focus area? (technical / product / general)
• Push frequency? (daily / weekly / real-time)
• Push time? (default 09:00)

Step 2: Write Config File

Path: ~/.ai-info-radar/push-config.json

{
  "focus": "technical",
  "sources": ["Jiqizhixin", "Liangziwei", "Andrej Karpathy", "Lex Fridman Podcast"],
  "push_time": "09:00",
  "frequency": "daily"
}

Step 3: Configure Cron (⚠️ Use UTC time; Beijing 09:00 = UTC 01:00)

# Example: configure your cron scheduler to run at UTC 01:00 daily
# The job should: read ai-info-radar push config, fetch latest content, format as daily digest, send to user

Step 4: Add heartbeat fallback

Configure a fallback check: if current time is 08:45-09:30 and today's digest hasn't been sent, trigger immediately.

Fetch Execution Logic

When triggered:

1. Read ~/.ai-info-radar/push-config.json
2. For each subscribed source, fetch latest content using appropriate tools
3. Filter: only content from past 24 hours
4. Format as daily digest (title + one-line summary + link)
5. Limit to 10 items max to avoid excessive length
6. Send to user via configured channel

Daily Digest Format

🤖 AI Daily Digest · {date}

📰 Today's Picks ({N} items):

1. **{title}**
   Source: {media} | {time}
   Summary: {one-liner}
   🔗 {link}

...

💡 Today's Highlight: {most important item}

---

Scheduled Push Configuration

Frequency Options

• Hourly, Every 3 hours, Every 6 hours
• Daily 9 AM, Daily 6 PM
• Monday, Friday

Topic Options

• All, Agent, LLM, Multimodal
• AI Safety, AI Products, AI Engineering, AI Startups

Count Options

• Per push: 1 item, 3 items, 5 items

Cron Configuration (⚠️ UTC Conversion)

# Example: Daily 9 AM push for Agent news, 3 items each time
# Note: cron uses UTC time. Beijing 09:00 = UTC 01:00
# Schedule: "0 1 * * *" → "Fetch latest Agent news and push to user, 3 items"

⚠️ UTC Conversion Reminder: Beijing time - 8h = UTC time. E.g., Beijing 09:00 → UTC 01:00, Beijing 18:00 → UTC 10:00.

Config Storage (V3)

User push config saved in ~/.ai-info-radar/push-config.json:

• Shareable across sessions
• Manually editable
• Import/export supported

---

Output Format

Card-style output, each item contains:

📰 [Media Name] · ⏱️ Published Time
🏷️ Content Tags
📝 Article Title
Summary (2-3 lines)
🔗 Original Link

---

Error Handling

Error Scenario	Handling
WeChat fetch failed (proxy issue)	Auto-fallback to web search, notify "content from third-party aggregation"
Twitter/X fetch failed	Retry once through ContentFetcher.fetch_from_twitter(<handle>), then web search fallback; after 3 consecutive failures, notify user
Website RSS broken	Switch to web_fetch for homepage article list
Push job failed	Add heartbeat fallback trigger rule

---

Gotchas

⚠️ WeChat Official Accounts may be blocked from direct access → Set HTTP_PROXY environment variable if needed

⚠️ Twitter/X no auth-token needed → scripts/content_fetcher.py uses xreach first, then guest-token GraphQL for login-free recent timeline fetch. For richer profile/tweet operations, use the dedicated x-twitter-scraper skill. Search is still limited without login; do not request auth tokens.

⚠️ RSS source becomes invalid → First try web_fetch for homepage articles, then fallback to web search

⚠️ Cron uses UTC time — don't enter Beijing time directly. Beijing -8h = UTC

⚠️ Cron push fails silently → Add heartbeat fallback to ensure daily digest isn't missed

⚠️ feedparser not pre-installed → Run pip install feedparser before first use

---

Hard Stop

If the same tool call fails more than 3 times, stop immediately. List all failed approaches and reasons, mark "Manual intervention required", and wait for user confirmation.

---

Dependencies

• Python 3.8+ (usually pre-installed)
• feedparser: pip install feedparser (RSS parsing; auto-prompted on first use)
• requests: pip install requests (usually pre-installed)

Recommended: agent-reach (enhanced search & fetch)

agent-reach provides superior multi-platform search (xreach), content extraction (xread), and API wrappers (mcporter/Exa). Without it, the skill falls back to Jina Search API and guest token GraphQL.

Install (choose one):

# Option 1: pip (recommended)
pip install agent-reach
agent-reach install --env=auto --safe

# Option 2: ClawHub
npx clawhub install agent-reach

After installation, run agent-reach doctor to verify. Optional: npm i -g mcporter && mcporter config add exa https://mcp.exa.ai/mcp for Exa semantic search.

Without agent-reach, the skill auto-falls back to Jina Search API + guest token GraphQL — still functional for most use cases.

---

Changelog

8.0.0 (2026-04-08)

• Added Xiaoyuzhou podcast SOP (iTunes RSS + faster-whisper), new references/podcast-xiaoyuzhou.md
• Added podcast row in Step 2 tool table

7.0.0 (2026-04-08)

• Added Weibo as data source: Chinese AI influencer tracking via web search + Playwright (visitor cookie)
• New references/weibo-source.md (scraping SOP + known AI influencer accounts)

6.0.0 (2026-04-08)

• Integrated X/Twitter login-free fetching through scripts/content_fetcher.py (xreach + guest-token fallback, no auth-token needed)

5.0.0 (2026-04-08)

• Cleaned up description triggers, removed unrelated entries
• New references/resource-list.md

3.0.0 (2026-04-08)

• Upgraded to real subscription + scheduled push system (cron-driven, auto content fetching)
• New "Subscription + Push Flow" section

2.0.0 (2026-04-07)

• Added execution flow (Step 1-3), platform tool selection and fallback rules
• Added error handling table

1.0.0 (Initial)

• Resource query, content fetching, smart recommendations, bookmarks, scheduled push
• 50+ AI media/blog/podcast/book resources aggregated