dailynews

Overview

This is a fully self-contained skill. It uses:

• A bundled Python RSS fetcher (fetch.py in the same directory as this file)
• Claude's native WebFetch tool for any supplementary scraping
• No gstack, no external browser dependency

Output: A ranked Vietnamese digest (~10 min read), saved to ~/.dailynews/YYYY-MM-DD.md and printed to the conversation.

---

Step 1 — Setup

TODAY=$(date +%Y-%m-%d)
SKILL_DIR=$(dirname "$(find ~/.claude -path '*/skills/dailynews/SKILL.md' 2>/dev/null | head -1)")
FETCH_SCRIPT="$SKILL_DIR/fetch.py"
DIGEST_DIR="$HOME/.dailynews"
DIGEST_FILE="$DIGEST_DIR/$TODAY.md"
mkdir -p "$DIGEST_DIR"

echo "TODAY: $TODAY"
echo "SKILL_DIR: $SKILL_DIR"
echo "FETCH_SCRIPT: $FETCH_SCRIPT"
echo "DIGEST_FILE: $DIGEST_FILE"

# Check if today's digest already exists
[ -f "$DIGEST_FILE" ] && echo "CACHED: yes" || echo "CACHED: no"

If CACHED is yes: Ask the user:

• A) Show cached digest (instant)
• B) Re-fetch fresh news (takes ~1-2 min)

Default to A unless they explicitly choose B.

---

Step 2 — Fetch RSS feeds

Run the bundled fetcher:

python3 "$FETCH_SCRIPT" 2>&1 | tee /tmp/dailynews_raw.json

If Python fails, try:

python "$FETCH_SCRIPT" 2>&1 | tee /tmp/dailynews_raw.json

Read the output:

cat /tmp/dailynews_raw.json

The output is a JSON object with:

• total_raw — total headlines scraped before dedup
• total_unique — unique stories after dedup
• articles — array of deduplicated story objects, each with:
  • title — canonical headline
  • sources — array of [source_name, url] pairs
  • source_count — how many sources covered this story
  • category_hint — rough hint: tech, pm, market, or design
  • pub_date — publication date string
• errors — sources that failed

If fewer than 3 sources responded, warn the user: "Chỉ có N nguồn phản hồi — kết quả có thể không đầy đủ. Tiếp tục?" and wait for confirmation.

---

Step 3 — Supplement with WebFetch for sources not in RSS

Use the WebFetch tool to grab headlines from these sources that are not in the RSS list:

Source	URL
Dribbble Popular	https://dribbble.com/shots/popular
Nielsen Norman Articles	https://www.nngroup.com/articles/
Daily.dev	https://app.daily.dev (skip if auth wall)

For each, call WebFetch with the URL and extract visible article titles and links from the page content. Add them to your working article list with appropriate category hints:

• Dribbble → design
• NNGroup → design
• Daily.dev → tech

---

Step 4 — Categorize all stories

For each article in your working list, assign exactly one of the 4 categories below based on title keywords and category_hint. When in doubt, use the primary subject of the headline.

Category A — 🚀 Xu hướng Công nghệ

What: AI models/products, new tech launches, SaaS releases, developer tools, funding rounds, acquisitions, open-source releases, platform updates, new products getting traction.

Keywords to look for: AI, LLM, model, launch, release, raises, funding, acquires, SaaS, API, open-source, developer, cloud, startup, GPT, Claude, Gemini, agent, automation

Category B — 🧠 Tư duy Quản trị Sản phẩm

What: PM tactics, product strategy, feature frameworks, A/B testing, growth loops, product-led growth, roadmap methods, team structures, lessons learned, best practices for building products.

Keywords: product manager, roadmap, strategy, framework, growth, retention, onboarding, feature, sprint, prioritization, lesson, best practice, how to build, product thinking, OKR, metrics

Category C — 📊 Phân tích Thị trường & Đối thủ

What: Big tech strategy moves, market share shifts, earnings reports, consumer behavior data, global mobile usage stats, M&A activity, regulatory/antitrust news, economic indicators affecting tech.

Keywords: market share, revenue, earnings, quarterly, billion, Apple, Google, Meta, Microsoft, Amazon, regulation, antitrust, download, DAU, MAU, users, global, report, study, data

Category D — 🎨 Thiết kế & Trải nghiệm (UX/UI)

What: New UI design patterns, UX research, accessibility, design system updates, typography/motion trends, usability studies, Figma updates, design tool news.

Keywords: UX, UI, design, usability, accessibility, Figma, prototype, user research, interface, design system, typography, animation, interaction, visual design, color

---

Step 5 — Rank within each category

Within each category, sort stories by:

1. source_count descending — more sources = higher rank
2. Tie-break: recency (more recent pub_date wins)
3. Stories from Stratechery, MIT Tech Review, HBR, NNGroup treated as authoritative single-source = weight 2

Assign tier badge:

• 🔥 Hot — 3+ sources
• 📌 Notable — 2 sources OR single authoritative source
• 💡 Worth Reading — 1 regular source

---

Step 6 — Write the digest

Produce the full digest in this format. All narrative text in Vietnamese. Keep article titles in their original English (add a short Vietnamese gloss in parentheses when the title is jargon-heavy).

# Tổng Hợp Tin Công Nghệ — {TODAY}
*Thời gian đọc: ~10 phút · {total_unique} tin từ {N} nguồn*

---

## 🚀 Xu hướng Công nghệ

**{tier_badge} {title}**
> {1-2 câu tóm tắt tiếng Việt — nêu điểm chính và tại sao quan trọng, max 50 từ}
🔗 {source_1}: {url_1}  {source_2}: {url_2}  ...

[repeat for each story in this category]

---

## 🧠 Tư duy Quản trị Sản phẩm

[same format]

---

## 📊 Phân tích Thị trường & Đối thủ

[same format]

---

## 🎨 Thiết kế & Trải nghiệm (UX/UI)

[same format]

---

## 📦 Sản phẩm mới hôm nay (Product Hunt)
*Top 5 theo upvotes*

1. **{product}** — {mô tả 1 câu tiếng Việt} 🔗 {url}
2. ...

---

*Nguồn thành công: {list} · Lỗi: {error list}*
*Cập nhật lúc: {timestamp}*

Vietnamese summary rules:

• 1-2 sentences per story, max 50 words
• Start with the most important fact, not "Theo bài viết..."
• Keep company names, product names, and numbers in original form
• For funding: always include amount + stage + investor if available
• For product launches: say what it does plainly
• For research: say the finding, not "a study found that..."

---

Step 7 — Save locally and output

Use the Write tool to save the digest to $DIGEST_FILE.

Then print the full digest to the conversation.

---

Step 8 — Save to Notion

After printing the digest, push it to Notion automatically.

8a — Generate the page name

The page title must follow the format: dd/mm/yy - {name}

Where {name} is a short (3-6 word) English descriptor of the day's top story or dominant theme. Examples:

• 03/04/25 - GPT-5 Launch & Apple Earnings
• 03/04/25 - AI Agents Week
• 03/04/25 - Meta Threads Hits 200M

Pick the name from the #1 ranked story across all categories (the one with the highest source_count). If two stories tie, pick the most impactful one.

Format the date as dd/mm/yy using the TODAY variable:

NOTION_DATE=$(date +%d/%m/%y)
echo "NOTION_DATE: $NOTION_DATE"

8b — Find the "Daily news" parent page

Use mcp__claude_ai_Notion__notion-search with query "Daily news".

Look for a page whose title is exactly or closely matches "Daily news". Extract its id as PARENT_PAGE_ID.

If no match is found, ask the user:

"Không tìm thấy trang 'Daily news' trong Notion. Bạn có muốn tôi tạo mới trang đó không, hay cung cấp ID trang parent?"

8c — Convert digest to Notion blocks

Map the digest markdown to Notion block format:

Markdown	Notion block type
# heading	heading_1
## heading	heading_2
**bold text**	paragraph with bold annotation
> blockquote	quote
🔗 Source: url	paragraph with link
---	divider
numbered list	numbered_list_item
bullet - item	bulleted_list_item
plain paragraph	paragraph

Build the children block array for the Notion page creation call.

Important: Notion API has a limit of 100 blocks per request. If the digest exceeds 100 blocks, split into multiple notion-create-pages or use notion-update-page calls. In practice, keep blocks under 95 to be safe.

8d — Create the child page

Call mcp__claude_ai_Notion__notion-create-pages with:

{
  "parent": { "page_id": "{PARENT_PAGE_ID}" },
  "properties": {
    "title": [{ "text": { "content": "{NOTION_DATE} - {name}" } }]
  },
  "children": [ ... blocks array ... ]
}

8e — Confirm success

After the page is created, print:

✅ Đã lưu lên Notion: "{NOTION_DATE} - {name}"
🔗 {notion_page_url}

If Notion creation fails:

• Print the error
• Tell user: "Bản digest đã được lưu cục bộ tại ~/.dailynews/{TODAY}.md"
• Do NOT retry automatically — let user decide

---

Step 9 — Final stats line

---
📈 {total_raw} headlines scraped → {total_unique} unique stories → {categorized} tiêu đề được xếp loại
📰 {N} nguồn thành công / 15+ · Lỗi/auth: {error_sources}
💾 Local: ~/.dailynews/{TODAY}.md
📓 Notion: {NOTION_DATE} - {name}

---

Adding custom sources

To add a new RSS source, edit ~/.claude/skills/dailynews/fetch.py and add a tuple to the SOURCES list:

("Source Name", "https://example.com/feed.rss", "rss", "tech|pm|market|design"),

To add Datanleo or any custom source with a known RSS URL, add it to SOURCES in fetch.py.

---

Error handling

Error	Action
Python not found	Tell user to install Python 3: brew install python3
Source returns 403/401	Mark [AUTH_REQUIRED], skip
Source returns 5xx	Mark [UNAVAILABLE], skip
JSON parse failure	Re-run fetch once, then skip if still failing
0 articles total	Tell user all sources failed, suggest checking network
Cached digest exists	Ask A/B (show cached vs re-fetch)

---

Important rules

• Never fabricate news. Only use what was actually fetched.
• Every story must have at least one real URL.
• All prose output in Vietnamese. Titles stay in English.
• If a source is a paywall and only the title is visible, include it and mark [paywall].
• Output goes directly to the conversation. Only the .md file is written to disk.
• Keep total digest length manageable: aim for 20-35 stories across all categories.