A Claude Code skill that generates a high signal-to-noise AI technical briefing from 60+ sources.
One command. 3 minutes. A structured 7-section briefing that tells you what actually matters.
Fetches from 3 source types in parallel:
| Strategy | Sources | Count |
|---|---|---|
| RSS Feeds | OpenAI, DeepMind, Anthropic, HN, LangChain, Karpathy, Stratechery, etc. | 18 feeds |
| Twitter/X | @sama, @DarioAmodei, @karpathy, @swyx, @a16z, etc. | 40+ accounts |
| WebSearch | Sources without RSS or Twitter (e.g., SSI) | fallback |
Outputs a structured technical briefing with 7 sections:
| # | Section | What's in it |
|---|---|---|
| 1 | Top Signals | 5 most important developments — each with [Fact], [Opinion], [Engineering Implication] |
| 2 | Builder's Changelog | New models, API changes, pricing, open-source releases |
| 3 | Engineer / Researcher Views | Core technical judgments on architectures, agents, reasoning, evals |
| 4 | Founder / Entrepreneur Views | Product opportunities, vertical AI, new application patterns |
| 5 | VC / Investment Signals | What investment narratives reveal about technology path validation |
| 6 | Architect Decision Lens | 4 questions every system architect should ask today |
| 7 | One Pattern to Remember | The single AI engineering pattern worth internalizing |
SKILL.md— Chinese (default)SKILL.en.md— English
Both versions share the same references/news-sources.md source configuration.
# Clone the repo
git clone https://github.com/MIAJIA/ai-news-skill.git
# Symlink to Claude Code skills directory
ln -s /path/to/ai-news-skill ~/.claude/skills/newsThen run /news in Claude Code.
For higher quality Twitter data, set up the Twitter MCP server. Without it, the skill falls back to WebSearch (lower signal).
ai-news-skill/
├── SKILL.md # Skill logic (Chinese)
├── SKILL.en.md # Skill logic (English)
├── README.md
└── references/
└── news-sources.md # Source configuration (RSS + Twitter + WebSearch)
Edit references/news-sources.md to add or remove sources. The file has 4 sections:
- RSS Feeds — add any RSS/Atom URL
- Twitter/X Accounts — accounts to track via Twitter MCP or WebSearch
- Non-RSS Sources — sources scraped via WebFetch or WebSearch
- Curation Rules — priority order, daily quota, dedup rules
- Lab blog auto-include: Posts from OpenAI, Anthropic, DeepMind, Google Research, and Meta AI are always included and never compete with HN for quota
- 7-day window for lab blogs: These publish infrequently; a 48h window misses most posts
- Twitter MCP > WebSearch:
site:x.comsearches mostly return profile pages, not tweets. Twitter MCP returns actual content with engagement metrics - 2-batch Twitter execution: 4 parallel Twitter API calls hit rate limits; 2 batches of 2 avoids this
- Fact/Opinion/Implication structure: Prevents the briefing from being a link dump
PRs welcome! Especially for:
- Adding high-quality information sources to
news-sources.md - Improving curation rules
- Adding support for new source types (e.g., Bluesky, Mastodon, YouTube transcripts)
To add a source, edit references/news-sources.md and submit a PR.
MIT