News Briefing Skill

Overview

This skill encapsulates a complete Twitter/X news briefing automation system that:

• Fetches posts from configurable Twitter/X accounts across multiple categories
• Generates star-rated, mobile-friendly briefings (readable in 20 seconds)
• Delivers briefings via Feishu/Lark messaging
• Deduplicates seen content with 7-day auto-cleanup
• Runs on a schedule (e.g., every 2 hours from 10:00 to 20:00), sending only once per day

System Architecture

News Briefing System
├── Data Layer: Fetch tweets from configured accounts (last 24h)
├── Dedup Layer: Skip already-seen content by tweet_id
├── Briefing Layer: AI generates star-rated summary (max 12 items)
├── Delivery Layer: Send via Feishu/Lark bot
├── Anti-repeat Layer: Daily sent_log prevents duplicate sends
└── Schedule Layer: Runs at intervals, exits early if already sent

Project Structure

project-directory/
├── config/sources.yaml          # Information source config (blogger list)
├── scripts/fetch_tweets.py      # Scweet tweet fetching script
├── data/sent_log.json           # Send status tracking
├── data/seen_tweets.json        # Seen content dedup (by tweet_id)
├── data/tweets_raw.json         # Raw fetched tweet data
└── .workbuddy/memory/           # Project memory

Setup Guide

Step 1: Configure Information Sources

Create config/sources.yaml with your selected accounts organized by category:

ICT Trading:
  - handle: example_trader1
    name: Example Trader 1
    followers: "100K"

US Stocks:
  - handle: example_trader2
    name: Example Trader 2
    followers: "200K"

AI:
  - handle: example_ai1
    name: Example AI Researcher
    followers: "150K"

Step 2: Set Up Tweet Fetching

Install Scweet for precise Twitter/X data fetching:

pip install scweet

Create scripts/fetch_tweets.py (see scripts/fetch_tweets.py in this skill for the full template). Key configuration:

• AUTH_TOKEN: Your Twitter/X auth_token (get from browser: login to x.com, F12, Application, Cookies, auth_token)
• ACCOUNTS: Dictionary mapping categories to lists of handles
• Requires VPN/proxy if accessing from regions where x.com is restricted

Step 3: Initialize State Files

Create data/sent_log.json:

{
  "last_sent_date": "",
  "last_sent_time": "",
  "history": []
}

Create data/seen_tweets.json:

{
  "items": [],
  "last_cleanup": ""
}

Step 4: Connect Feishu/Lark

1. Open WorkBuddy connector management
2. Find Feishu/Lark and connect/authorize
3. Verify connection with: lark-cli auth status --json --verify
4. Get your user open_id from the auth status output

Step 5: Create Automation

Create a recurring automation in WorkBuddy with:

• Schedule: e.g., FREQ=DAILY;BYHOUR=10,12,14,16,18,20;BYMINUTE=0
• Connectors: feishu
• Working directory: Your project path

The automation prompt should follow the six-step workflow below.

Six-Step Workflow

Step 1: Check Daily Send Status

Read data/sent_log.json. If last_sent_date equals today, output "Already sent today, exiting" and stop immediately.

Step 2: Fetch Last 24h Tweets

Option A: Scweet Precise Fetching (requires VPN/proxy)

python3 scripts/fetch_tweets.py --hours 24 --limit 5 --output data/tweets_raw.json

• Success (non-empty array) -> Use ONLY Scweet data, add VPN emoji to title
• Failure/empty array -> Fall back to Option B

Option B: WebSearch Fallback (VPN not connected)

Search each category's accounts using WebSearch. Collect results from the last 24 hours.

VPN Status Marker: If Scweet succeeds, add a VPN emoji to the briefing title. If WebSearch fallback is used, do not add the emoji.

Step 3: Deduplication

Read data/seen_tweets.json. Skip any tweets whose tweet_id matches existing entries.

Step 4: Generate Briefing

Importance Scoring (two factors combined)

1. Social impact of the information
2. Impact on US stock trading

• Both high -> 3 stars
• One high + one medium, or both medium -> 2 stars
• Both low -> 1 star

Format Requirements

• Maximum 12 items, readable in 20 seconds on mobile
• Grouped by star rating: 3 stars first, then 2 stars, then 1 star
• Each item numbered with emoji: 1️⃣ 2️⃣ 3️⃣ ... 🔟
• Each item format (hyphen-separated, no brackets shown): Category-@Blogger-CoreContent-TradingImpact-OtherSignificance
• Categories: ICT Trading / US Stocks / AI / Politics / Finance / Tech
• Core content: One-sentence summary (20-40 chars)
• Trading impact: e.g., "Highly impacts XXX stock" / "Low impact on XXX" / "No direct impact"
• Other significance: One-sentence supplementary insight (industry trend / forward judgment / risk alert)
• Language: Mixed (names/terms/tickers in English, analysis in user's preferred language)
• One blank line between each item for readability

Title Format (includes push time)

• VPN connected (Scweet success): 📰 Daily Briefing | {date} {HH:MM} VPN
• VPN not connected (WebSearch): 📰 Daily Briefing | {date} {HH:MM}

Footer Format

• Scweet success: — {N} items / Scweet / next update {time}
• WebSearch fallback: — {N} items / WebSearch / next update {time}

Important: When Scweet succeeds, use ONLY Scweet data. Do not mix with WebSearch results.

Format Example

📰 Daily Briefing | 2026-07-16 10:00 VPN

⭐⭐⭐

1️⃣ Finance-@example-Market drops 5% on rate fears-Highly impacts SPY/QQQ-Risk-off sentiment rising

2️⃣ Tech-@example-New AI chip announced-Low impact on sector-Competitive landscape shifting

⭐⭐

3️⃣ US Stocks-@example-ES futures grind higher-Highly impacts long strategies-Maintain long positions

⭐

4️⃣ AI-@example-New research paper published-No direct impact-Watch for commercialization timeline

— 4 items / Scweet / next update 12:00

If no valuable new content across all categories, output "No significant updates today" and end without marking as sent.

Step 5: Send via Feishu/Lark

lark-cli im +messages-send --user-id {YOUR_OPEN_ID} --as bot --text $'briefing content'

• Returns ok: true -> Success
• Failure -> Do not mark as sent, will retry next run

Step 6: Update State

After confirmed send success:

1. Update data/sent_log.json:

   • Set last_sent_date to today (YYYY-MM-DD)
   • Set last_sent_time to current time (ISO 8601)
   • Append to history array: {"date": "today", "time": "now", "preview": "first 50 chars"}

2. Update data/seen_tweets.json:

   • Add all briefing items with: {"tweet_id": "id", "username": "handle", "summary": "summary", "date": "today", "seen_at": "now"}
   • Clean up entries older than 7 days
   • Update last_cleanup to today

Important Rules

• Only execute Step 6 after Step 5 send success
• If any step fails, do not mark as sent; next run will retry
• Maximum one send per day; subsequent runs exit early
• When Scweet succeeds, use only Scweet data (no WebSearch mixing)
• VPN/proxy connection status is indicated by the VPN emoji in the title

Adding New Bloggers

Users can add new bloggers by saying in conversation: "Add {category} blogger {handle}"

The AI then automatically:

1. Adds the blogger to config/sources.yaml under the appropriate category
2. Adds the handle to scripts/fetch_tweets.py in the ACCOUNTS dictionary
3. Updates the automation prompt's account list

See references/add-blogger.md for detailed instructions.

Key Dependencies

• Scweet 5.3+: Python package for Twitter/X data fetching via auth_token
• Twitter/X auth_token: Obtain from browser cookies (requires periodic renewal)
• Feishu/Lark connector: Enabled in WorkBuddy connector management, bot identity sending
• VPN/Proxy: Required for Scweet to access x.com from restricted regions
• WorkBuddy Automation: For scheduled execution