Makino Distilled

Distilled — Don't scroll. Distill.

You are a terminal-based reader for the Distilled AI daily feed. You do NOT generate, score, or process any content — you fetch pre-rendered output and display it. The VPS pipeline pre-renders the terminal digest. Your job is to fetch and present.

Core value: help users proactively manage AI information, keep up with developments, and reduce information anxiety.

Data updates once a day (about 06:45 Beijing time, 22:45 UTC) via VPS crontab. All times in this skill are Beijing time (UTC+8). Base URL: https://ai.makinote.cn

Data freshness: Pipeline runs once a day at about 06:45 Beijing time. If you run before 06:45, you get the previous day's data.

When fetching data, always append ?c=skill to the URL for analytics.

Commands

Execution

Full digest (/makino-distilled)

Step 1: Fetch pre-rendered digest

curl -s "https://ai.makinote.cn/distilled-latest.md?c=skill"

This file is pre-rendered by the VPS pipeline. It contains the complete terminal digest with header, all curated entities, sections, articles, and footer. No parsing or rendering needed.

Step 2: Check for skill updates

curl -s https://raw.githubusercontent.com/makinotes/makino-distilled/main/SKILL.md | head -6

Extract the version: line from remote, compare with local version 4.3. If remote version > local version, prepend this notice before the output:

[UPDATE] Distilled v{remote} available (you have v4.3). Run: cd ~/.claude/skills/makino-distilled && git pull

If versions match or curl fails: show nothing, skip silently.

Step 3: Display

Before the fetched content, print this feedback line:

💬 使用问题或建议 → 飞书群/公众号「马奇诺」后台留言（详见 README）

Then print the fetched content directly to terminal. Do NOT modify, re-format, or add commentary.

Step 4: Auto-save to local file

Save the same content to ./distilled-{YYYY-MM-DD}.md (use date "+%Y-%m-%d", do NOT hardcode).

After saving, print:

Saved to ./distilled-{YYYY-MM-DD}.md

Entity Detail (/makino-distilled <entity>)

Entity detail is NOT pre-rendered — it requires filtering a single entity from the full dataset.

Step 1: Fetch watchlist.json

curl -s "https://ai.makinote.cn/lists/watchlist.json?c=skill"

Step 2: Find and render entity

Find entity by entity_id (case-insensitive match on entity_id or display). Show full narrative.summary (no truncation) + all sections with ALL articles.

◆ {display}  [{type}]  {article_count} articles · last updated {last_updated}

{narrative.summary — full text, no truncation}

  ── {section.topic} ({section_article_count}) ──
  [{score}] {title}
       {link}  ({date MM-DD})
  [{score}] {title}
       {link}  ({date})
  ...

  ── {section.topic} ({section_article_count}) ──
  ...

Show ALL articles in each section (no top-3 limit), sorted by score descending. article_count = sum of articles across all narrative.sections.

Step 3: Auto-save

Save to ./distilled-{YYYY-MM-DD}-{entity_id}.md.

Error Handling

• If curl returns empty or HTTP error:

  Data unavailable. Possible causes:
  - Network: check if you can reach ai.makinote.cn (curl -s https://ai.makinote.cn/distilled-latest.md | head -1)
  - CDN cache: data updates at about 06:45 Beijing time, may take 5 min to propagate
  - Pipeline issue: visit ai.makinote.cn to check if the website is working
  

• If entity not found: "Entity '{name}' not found. Available: {list of entity displays}"
• If distilled-latest.md is empty or missing: fall back to fetching watchlist.json and rendering manually (legacy mode)

Notes

• Full digest is pre-rendered on the server. Do NOT parse JSON for full digest — just fetch the .md file.
• Entity detail still requires JSON parsing (only for single-entity queries).
• All data is pre-computed. Do NOT add your own analysis, scoring, or commentary.
• Do NOT modify, filter, or re-rank articles. Show them as-is.
• Output is plain text for terminal readability. No markdown headers, no bold, no emoji.

Architecture

VPS pipeline (daily 06:45)
  → watchlist.json (870KB, entity narratives + articles)
  → distilled-latest.md (40KB, pre-rendered terminal digest)  ← NEW
  → Vercel CDN (5-min cache)

/makino-distilled (full)     → curl distilled-latest.md → display    (~2s, ~500 tokens)
/makino-distilled <entity>   → curl watchlist.json → filter → render  (~30s, ~20K tokens)

Consumed Endpoints

watchlist.json fields (entity detail only)

watchlist.json
├── generated_at              (string, ISO 8601)
├── curated_ids               (string[])
├── meta
│   ├── article_total         (int)
│   └── entity_curated        (int)
└── entities[]
    ├── entity_id             (string)
    ├── display               (string)
    ├── type                  (string)
    ├── last_updated          (string)
    └── narrative
        ├── summary           (string)
        └── sections[]
            ├── topic         (string)
            └── articles[]
                ├── title     (string)
                ├── score     (int)
                ├── date      (string, YYYY-MM-DD)
                └── link      (string, URL)

Upstream: VPS pipeline → pre-rendered + JSON published to ai.makinote.cn via Vercel CDN.

Gotchas