makino-distilled — Don't scroll. Distill.

130+ AI 信源按主题聚合评分，只看高质量信息，追踪长期趋势不追热点，降低信息焦虑。

Install

cd ~/.claude/skills/
git clone https://github.com/makinotes/makino-distilled.git

Then type /makino-distilled in Claude Code.

Usage

/makino-distilled

Full digest — every curated entity at a glance (per-section top 3 articles).

/makino-distilled claude
/makino-distilled openai
/makino-distilled agent

Single entity deep dive — full narrative + all articles.

Output auto-saves to ./distilled-{date}.md.

Example Output

Full output from a real run (752 lines, 24 entities, 969 articles):

demo/distilled-2026-04-03.md

Preview (first 30 lines):

DISTILLED · 04-03
Don't scroll. Distill.
Proactively manage AI info · Keep up with developments · Reduce anxiety

969 articles · 24 entities · data from 04-03 16:57

━━━ WATCHING ━━━

◆ Claude  [Product]  33 articles
  最近最值得关注的是 Claude Code 源码泄露和从中暴露的 Harness Engineering 工程实践...

  ── Claude Code 源码泄露 (6) ──
  [77] 突发！Claude Code 源码泄露，扒出这些隐藏功能
       https://mp.weixin.qq.com/s/...  (04-01)
  [72] Claude Code 512,000 行源码泄露，有3点启发
       https://mp.weixin.qq.com/s/...  (03-31)

  ── Harness (6) ──
  [70] 一夜之间，全世界的 Agent 能力提高了一个档次
       https://mp.weixin.qq.com/s/...  (04-03)

  ...24 entities, ~750 lines total

Configuration

There is nothing to configure. Every user sees the same curated ranking; scoring and entity tracking live in the backend. If a dimension matters to you, open an issue and say which one.

Features

Zero dependencies, zero API keys (just curl)
Data refreshes once a day, around 06:45 Beijing time (22:45 UTC)
Auto-detect data freshness (stale data warning)
Auto-check for skill updates on each run
Same data as ai.makinote.cn, terminal-native

What Problem Does This Solve

Keeping up with AI developments means scrolling through dozens of sources daily. Most of it is noise. This skill distills 130+ sources into structured, scored summaries — you read the terminal output in 2 minutes instead of scrolling feeds for an hour. Know what matters, skip what doesn't.

Data

All data is fetched from public JSON endpoints at ai.makinote.cn.

The pipeline runs once a day on a VPS via crontab, at about 06:45 Beijing time (22:45 UTC).

If you run before the daily update, you get the previous day's data. The skill shows a freshness note when data is stale.

Endpoint	Content
`lists/watchlist.json`	Entity narratives + articles (30-day window)

How It Works

130+ sources --> Pipeline (VPS) --> JSON API (CDN) --> This skill (curl + render)

The pipeline fetches, scores, summarizes, and structures articles once a day.
This skill is a read-only terminal client — it fetches the JSON and formats it.
No local processing, no LLM calls, no state.

For Developers

The JSON endpoints are public. Build your own client:

# Entity narrative
curl -s https://ai.makinote.cn/lists/watchlist.json | \
  jq '.entities[] | select(.entity_id == "claude") | .narrative.summary'

# All curated entity IDs
curl -s https://ai.makinote.cn/lists/watchlist.json | jq '.curated_ids'

Website

ai.makinote.cn — same data, visual interface.

Update

cd ~/.claude/skills/makino-distilled && git pull

The skill also checks for updates automatically on each run. If a new version is available, you'll see a notice in the output.

FAQ

Q: /makino-distilled no response or error?
Make sure you cloned into the correct directory (~/.claude/skills/makino-distilled). The folder name must be exactly makino-distilled. If you see curl errors, check if your network can reach ai.makinote.cn.

Q: What are the 130+ sources?
Chinese and English AI media, tech blogs, research labs, newsletters, and developer communities. The full source list is curated and maintained on the backend. You can browse entity coverage at ai.makinote.cn.

Q: Who picks the tracked entities? Can I add my own?
The curated entities (Claude, Agent, OpenAI, etc.) are picked by the pipeline maintainer. Custom entity tracking is not supported yet — this is a read-only client.

Q: What does the score (e.g. [83]) mean?
A relevance score from 0-100 assigned by the pipeline. Higher = more relevant to the entity topic. It is NOT a quality rating of the article itself.

Q: What is the "30-day window"?
The pipeline tracks articles from the past 30 days. Older articles roll off automatically. This keeps the digest focused on recent developments.

Q: I see [NOTE] Data is from yesterday — is something broken?
No. The pipeline updates once a day at about 06:45 Beijing time. If you run before that, you get the previous day's data. This is normal.

Q: Output is very long. Can I see just one entity?
Yes. Use /makino-distilled claude or /makino-distilled agent to deep dive into a single entity with full article lists.

Q: The .md files keep piling up. Should I clean them?
The skill saves one file per day (distilled-YYYY-MM-DD.md) in your working directory. Delete old ones whenever you want — they are just local snapshots.

Q: How do I update the skill?
Run cd ~/.claude/skills/makino-distilled && git pull. The skill also auto-checks for updates on each run and shows a notice if a new version is available.

Q: Will git pull overwrite my changes?
If you edited SKILL.md locally, git pull may conflict. Recommendation: don't modify SKILL.md — open an issue on GitHub instead.

Q: Is this the same as the website?
Same data source, same JSON. The website has visual design and interactive features. The skill is a terminal-native reader optimized for agentic workflows.

Q: Will this project be maintained?
Yes. The pipeline runs automatically, and this skill is updated whenever the upstream JSON changes.

Q: Are article links guaranteed to work?
Links are scraped from original sources. Some may expire or get paywalled over time. The pipeline does not archive article content.

License

Apache 2.0 — see LICENSE and NOTICE

Community & Contact

Bugs, questions and ideas: open an issue. 也可以在公众号「马奇诺」后台留言。

Author

makino