メニュー
Personalized top news briefing with trend insights. Use when the user asks for news, briefings, what's trending, hot topics, daily digest, or 新闻/热点/快报. Supports multiple categories, scheduled delivery, and preference learning.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
SOC 職業分類に基づく
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/fang-lin/skills --skill top-newsコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
| name | top-news |
| description | Personalized top news briefing with trend insights. Use when the user asks for news, briefings, what's trending, hot topics, daily digest, or 新闻/热点/快报. Supports multiple categories, scheduled delivery, and preference learning. |
| version | 0.4.1 |
| license | MIT |
| compatibility | Python 3.9+, internet access required. No paid API keys needed for basic operation (NewsAPI optional). |
| platforms | ["macos","linux","windows"] |
| required_environment_variables | [{"name":"NEWSAPI_KEY","prompt":"NewsAPI key (optional — leave blank for free RSS-only mode)","help":"https://newsapi.org/register","required_for":"expanded news coverage beyond free RSS feeds"}] |
| metadata | {"author":"fang-lin","repo":"https://github.com/fang-lin/top-news","hermes":{"tags":["news","briefing","rss","trending","digest","personalized","新闻","热点"],"category":"productivity","requires_toolsets":["code_execution","web","skills","memory","cronjob"],"config":["[Truncated]","[Truncated]","[Truncated]","[Truncated]","[Truncated]"]}} |
Deliver personalized top news briefings with trend insights, preference tracking, and scheduled delivery.
[!IMPORTANT]
- Use the scripts for data fetching and ranking. Do NOT scrape news yourself. Use
fetch_news.pyfor data collection andrank_and_dedup.pyfor ranking.- YOUR job is prompt composition, trend analysis, and user interaction. The scripts handle data; you handle intelligence.
- Respect user preferences. Always check the config before delivering.
- Never fabricate news. Only report what the scripts return. If data is insufficient, say so.
- Send briefings as text messages. Not files, not images. Clean formatted text.
- Use the user's language. Detect from conversation context and match it. All onboarding prompts, briefings, and interactions should be in the user's language. This SKILL.md is in English for portability — you translate at runtime.
- Follow the exact formatting in templates. Onboarding questions use plain text with aligned spacing — NOT markdown tables. Copy the format exactly, only translate the text. Do not reformat into tables, bullets, or any other structure.
- Don't patch skill files yourself. If you find a bug or missing feature in this skill, tell the user and suggest running
hermes skills update top-news. The skill is maintained upstream — local edits will be overwritten on the next update.
[!IMPORTANT] Onboarding MUST be multi-turn. Ask ONE question at a time. Wait for the user's answer before asking the next. Do NOT dump all questions at once. After all questions, present a summary for review. Only proceed after explicit approval.
Option formatting rules:
- Every option MUST have a number prefix (1, 2, 3...) for easy reply
- Each question MUST state whether it is multi-select or single-select
- Multi-select questions: mark each option with ✅ (selected) or ❌ (not selected). First-time: all enabled by default. Returning user: show current selections. User replies with numbers to TOGGLE.
← default (first-time) or ← current (returning user).If no config exists yet, walk the user through setup step by step. If config already exists (user wants to change settings), show current values.
Question 1: Topics [MULTI-SELECT — reply with numbers to toggle]
What topics interest you? (reply with numbers to toggle on/off)
1. 🖥 Technology ✅
2. 🤖 AI/ML ✅
3. 💰 Finance ✅
4. 🌍 World News ✅
5. 🔬 Science ✅
6. 🎬 Entertainment ✅
All enabled by default. Reply with numbers to toggle (e.g. "5 6" to disable Science & Entertainment), or 0 to skip.
WAIT for answer.
Question 2: Custom RSS Feeds [OPTIONAL]
Want to add any custom RSS feeds? (in addition to Google News)
Enter RSS/Atom URLs (one per line), or 0 to skip.
WAIT for answer.
Question 3: Language [SINGLE-SELECT]
Language preference?
1. 🇨🇳 Chinese
2. 🇬🇧 English
3. 🌐 Both ← default
Reply 1/2/3, or 0 to skip.
WAIT for answer.
Question 4: Format [SINGLE-SELECT]
How detailed should each briefing be?
1. 📋 Headlines only — just titles and sources
2. 📝 With summaries — 2-3 sentence per item ← default
3. 📊 Deep analysis — detailed context and implications
Reply 1/2/3, or 0 to skip.
WAIT for answer.
Question 5: Item Count [SINGLE-SELECT]
How many news items per briefing?
1. 5
2. 10 ← default
3. 15
4. Custom number
Reply 1/2/3/4, or 0 to skip.
WAIT for answer.
Question 6: Timezone [SINGLE-SELECT]
[!IMPORTANT] Cron jobs run in UTC on the server. You MUST know the user's timezone to schedule deliveries at the correct local time. Try to auto-detect from conversation context (language, mentioned city, greeting time). If uncertain, ask explicitly.
What's your timezone?
1. 🇨🇳 Asia/Shanghai (UTC+8)
2. 🇺🇸 America/New_York (UTC-4)
3. 🇺🇸 America/Los_Angeles (UTC-7)
4. 🇪🇺 Europe/Berlin (UTC+2)
5. 🇬🇧 Europe/London (UTC+1)
6. 🌐 Other (tell me your city or UTC offset)
Reply 1-6, or 0 to skip (defaults to UTC).
WAIT for answer.
Question 7: Schedule [SINGLE-SELECT]
When should I deliver? (times in your local timezone)
1. 🌅 Morning (08:00) ← default
2. 🌆 Evening (20:00)
3. 🌅+🌆 Both (08:00 + 20:00)
4. ⏰ Custom time(s)
Reply 1/2/3/4, or 0 to skip.
WAIT for answer.
Question 8: Delivery Target [SINGLE-SELECT]
Where should I send the briefing?
1. 💬 Here (this chat) ← default
2. 📢 A specific group or channel
Reply 1/2, or 0 to skip.
WAIT for answer.
Review Summary
After all questions are answered, present a complete summary:
Present as a clean aligned summary. Use monospace or pre-formatted text to ensure alignment in Telegram:
Your briefing configuration:
📋 Topics [list of selected topics]
🔍 Source Google News (+ [N] custom RSS feeds)
🌐 Language [choice]
📝 Format [choice]
🔢 Items [count] per briefing
🕐 Timezone [timezone]
⏰ Schedule [times] (local) → [UTC times] (UTC)
💬 Delivery [target]
Confirm? Reply "yes" to apply, or tell me what to change.
NOTE: In Telegram, use code blocks or consistent spacing to keep the summary visually aligned. Avoid markdown tables — they don't render well in Telegram.
STOP and WAIT for explicit approval.
If the user says yes:
All set! Here's what I configured:
✅ Config saved
✅ [N] news sources activated
✅ Cron scheduled: [times]
Your first briefing will arrive at [next scheduled time].
Want me to run a test briefing right now?
All data lives in the agent's workspace:
workspace/top-news/
├── config.json # User preferences
├── sources.json # Active sources (generated from preferences)
├── tracking.json # 7-day rolling window for dedup + streaks
├── preferences.json # User feedback (👍👎) history
├── history/ # Past briefings
│ └── 2026-04-27.json
└── logs/
└── 2026-04-27.log
{
"interests": ["technology", "ai", "finance"],
"interest_weights": {"technology": 0.4, "ai": 0.4, "finance": 0.2},
"language": "both",
"format": "summary",
"item_count": 10,
"timezone": "Asia/Shanghai",
"schedule": "08:00",
"delivery_target": "current_chat"
}
Generated from user topics. See references/sources-catalog.md for query keyword mapping:
[
{"type": "google-news", "topic": "technology", "language": "en", "region": "US"},
{"type": "google-news", "topic": "AI artificial intelligence", "language": "en", "region": "US"},
{"type": "google-news", "topic": "finance stock market", "language": "en", "region": "US"}
]
For bilingual users (language = "both"), generate two entries per topic — one per language. Users can also add custom RSS feeds:
{"type": "rss", "name": "My Blog", "url": "https://example.com/feed/"}
Step 1: Fetch raw data
from hermes_tools import terminal
SCRIPT_DIR = "<skill scripts directory>"
WORKSPACE = "<agent workspace>/top-news"
result = terminal(
f'python {SCRIPT_DIR}/fetch_news.py '
f'--sources {WORKSPACE}/sources.json '
f'--output {WORKSPACE}/raw_news.json '
f'--limit 20'
)
print(result["output"])
Step 2: Rank and deduplicate
result = terminal(
f'python {SCRIPT_DIR}/rank_and_dedup.py '
f'--input {WORKSPACE}/raw_news.json '
f'--tracking {WORKSPACE}/tracking.json '
f'--output {WORKSPACE}/ranked_news.json '
f'--top 10'
)
print(result["output"])
Step 3: Compose briefing
Read ranked_news.json and compose the briefing text using the templates in templates/briefing-template.md. Choose format based on user config.
For each article:
Step 4: Add trend insights
Analyze the tracking.json to identify:
Write a brief trend section at the end.
Step 5: Deliver
Send the composed briefing as a text message to the user.
Step 6: Save history
Save the delivered briefing to history/YYYY-MM-DD.json.
Step 7: Collect feedback
After delivery, prompt:
💬 Reply with 👍3 or 👎5 to react, or tell me what you want more/less of!
When user replies with feedback, update preferences.json:
{
"likes": [{"title": "...", "source": "...", "date": "..."}],
"dislikes": [{"title": "...", "source": "...", "date": "..."}],
"adjustments": ["more AI news", "less entertainment"]
}
User can say things like:
Update config.json and sources.json accordingly.
[!IMPORTANT] Hermes cron runs in UTC. You MUST convert the user's local time to UTC before creating the cron schedule. Read
timezonefromconfig.jsonand compute the UTC offset.Example: User wants 08:00 in Asia/Shanghai (UTC+8) → cron should be
0 0 * * *(00:00 UTC).
Use Hermes cron to schedule delivery:
/cron add --name "top-news" --schedule "0 0 * * *" --prompt "Deliver my top news briefing"
Adjust the cron schedule based on user's preferred delivery time after converting to UTC.
references/sources-catalog.md — Available data sources by categorytemplates/briefing-template.md — Output format templates (headlines, summary, deep analysis)fetch_news.py returns fewer items than expected, check references/sources-catalog.md for known broken feeds and suggest alternatives.fetch_news.py returned >0 items and history/YYYY-MM-DD.json was saved.hermes cron list to confirm the job exists with the correct schedule.config.json to confirm the update persisted.