Think clearly, write naturally.
Claude Code skills that turn your CLI into a thinking + writing tool. Like gstack is for coding, Lumen is for thinking.
In the post-AI era, anyone can generate text. What's scarce is structured personal knowledge + judgment.
You consume YouTube videos, read articles, have conversations — but those fragments scatter across apps and vanish from memory. When it's time to write, you start from zero.
Lumen captures your knowledge fragments, indexes them automatically, and uses Socratic Q&A to help you think clearly — so writing emerges naturally from accumulated insight.
You never organize. You only judge what matters. AI does everything else.
- Claude Code with an active subscription (Pro or Max)
- macOS or Linux (Windows via WSL)
- Git
git clone https://github.com/howardpen9/lumen.git ~/.claude/skills/lumencd ~/.claude/skills/lumen && ./bin/setupYou should see:
✦ Lumen — Setting up skills
✓ Skills linked: ~/.claude/skills/lumen
✓ Data directory: ~/.lumen
✓ 8 skills installed
Ready! Try these in Claude Code:
/lumen-capture "your first thought"
/lumen-index
/lumen-think
Open Claude Code in any directory and run:
/lumen-capture "your first thought goes here"
That's it. You're capturing.
Open Claude Code and try:
/lumen-capture "I think AI agent routing is fundamentally a product design problem, not an engineering one"
Lumen saves it to your knowledge pool, auto-tags it, and connects it to related thoughts.
/lumen-youtube https://youtube.com/watch?v=dQw4w9WgXcQ
AI summarizes the video, extracts key insights, and asks you: which insights matter to you? Your judgment is the product.
/lumen-index
✦ Your Knowledge Pool
12 thoughts · 3 explorations · last capture: 2 hours ago
● AI Agent Routing Crystallizing
8 thoughts · 2 refs · evolving since Mar 19
→ Ready to think deeper? Try /lumen-think agent-routing
○ Context Engineering Exploring
3 thoughts · 1 paper · new since Mar 21
◌ Tauri vs Electron Seed
2 thoughts · waiting for more input
/lumen-think agent-routing
AI presents your accumulated thoughts, identifies tensions, and asks Socratic questions — one at a time:
"You keep saying 'context-aware routing.' What does that actually mean in practice?"
"You believe static rules are wrong, but your Agent Hub still uses them. Why?"
"If you had to explain your position in one sentence, what would it be?"
After 4-6 questions, AI suggests an outline based on your answers.
/lumen-draft agent-routing
Turns your thoughts + Q&A insights into a coherent draft — in your voice, using your words. AI connects and expands, but never fabricates.
/lumen-export agent-routing --format x-thread
/lumen-export agent-routing --format linkedin
/lumen-export agent-routing --format blog
| Skill | What it does | When to use |
|---|---|---|
/lumen-capture |
Capture a thought, URL, or YouTube video | Anytime — see something worth remembering |
/lumen-youtube |
Index a YouTube video — AI summary + key insights | Found a video but no time to watch it all |
/lumen-index |
View your knowledge pool, auto-organized | Want to see what you've been thinking about |
/lumen-think |
Socratic Q&A to deepen an exploration | Have enough material, need to clarify your position |
/lumen-draft |
Generate a draft from accumulated thoughts | Ready to write — thoughts have crystallized |
/lumen-polish |
Refine in your voice (zh/en style profiles) | Draft done, need language quality |
/lumen-export |
Format for X thread / LinkedIn / blog | Ready to publish to a specific platform |
/lumen-review |
Weekly learning review — see yourself think | End of week — reflect on what you learned |
Your thoughts naturally progress through stages — AI determines status automatically:
Skills can be used independently or in sequence. Each skill reads from and writes to ~/.lumen/.
Everything lives in ~/.lumen/ — local, portable, Git-friendly. No cloud, no account, no vendor lock-in.
~/.lumen/
├── pool/ # Raw captures (markdown + YAML frontmatter)
├── explorations/ # AI-grouped topic clusters (JSON)
├── drafts/ # Generated drafts (markdown)
├── exports/ # Platform-formatted outputs
├── memory/ # Your writing style profiles (zh/en)
├── youtube/ # Video summary cache
└── reviews/ # Weekly review snapshots
See DATA-FORMAT.md for the full specification.
Lumen doesn't write for you. It materializes your judgment.
Every time you say "this matters" or "I disagree with that" — that's a judgment call. Those calls are the atomic unit of knowledge work in 2026. AI can summarize, translate, and reformat — but only you can decide what's important.
Lumen captures those decisions, connects them over time, and helps you see what you actually think.
Lumen leverages your existing Claude Code subscription. No extra API keys. No extra cost. We don't sell AI — we help you use the AI you already have.
Inspired by:
- gstack — Skills that make Claude Code 10x productive for coding
- Pencil.dev — Design intent → AI → design output
- Conductor — Parallel AI agents, native macOS
"Writing is dead, notes are alive" — more precisely: pure writing skill is commoditized, but structured personal knowledge + judgment is more valuable than ever.
Lumen's output isn't articles. It's structured self-awareness of your own intellectual growth. Articles, threads, and posts are just one serialization format.
MIT