learnprompt/luban-skill
Deployment & DevOpsThis page was translated automatically. Technical details may differ from the original. View the original
Luban | Luban — Refines 'usable skills' into public assets that can be installed, disseminated, verified, and evolved. Agent skill-polishing workshop: material inspection, inspection, inspection, rulering, slow planing, and reworking
Overview
Claude Code users can also use the plugin marketplace (automatic updates): - Posting it on GitHub, no one installs it—others don't understand what it's for; - The README is like a project specification, with no first-screen hook and no product to take screenshots; - You say it "works well," but you can't provide any evidence to reproduce it; - Want to optimize, but don't know where to start — change the trigger word? Rewrite the workflow? Add a showcase? The usual approach is to "polish it for me." Luban's approach was to treat it as ** and store it in the workshop: first challenge whether it was worth carving, then see how competitors could stand, measure three rulers to identify shortcomings, cut and plan, each cut must pass the verification gate, and finally issue a block to set the rules. - : Material inspection conclusions, peer benchmarking table (all with URLs), niche judgment, scoring table, three polishing directions, directly replaceable rewrite segments - : Before and after polishing scores, one-sentence new positioning, special skills, next steps - : One-time comparison scripts solidified into tools, judgment standards established as explicit project rules - "Let Luban take a look at this skill" / "Refine my class door" - "Upgrade my skill" / "Polish my skill" - "Skill checkup" / "Skill audit" - "Why is no one installing my skill?" - "How to publish this skill to GitHub/ClawHub" - "Benchmark it against similar skills" Luban's first single task: polished ai-news-radar (about 1k stars) from v0.6 to v0.7.0, completed in one conversation, merged all 4 PRs: - **Identify data links silently stopped updating under all green Actions** (git add whitelist omitted) - for score repair **Historical data review verification: clear, zero accidental damage - Single source proportion in the featured area; First screen rendering, page height - verification methods accumulated into warehouse tool backtest_scoring.py, with the project rule "dynamic scoring must be accompanied by ≥14-day playback." Full process documentation (including incidents and lessons): skills/luban/examples/ai-news-radar-case.md—each number is accompanied by a clickable PR link.
Overview (machine translated)
🌐 English · Chinese
Luban | Luban
Bring your skills to the classroom door and let the master refine them.
**Refine a “usable skill” into a public asset that can be understood, installed, disseminated, validated, and continuously evolved. **
Installation · Five Actions · Combat Record · How Is It Different from Its Peers · Safety Boundaries · Acknowledgments
Get Started Quickly
Claude Code users can also use the plugin marketplace (automatic updates):
Hermes Agent User:
Or install manually:
After installing, he said to the agent:
Luban will first complete material inspection, inspection, positioning, and rulering, giving you three polishing directions and recommending one—Before you choose the direction, it won’t move a single line of your words.
What problems does it solve?
You wrote a skill, and it works well for yourself. And then?
- Posted on GitHub, no one installed it—others couldn’t understand what it was for;
- The README is like a project manual, with no first-screen hook and no product that can be screenshotted;
- You say it “works well,” but you can’t provide a reproachable evidence;
- Want to optimize, but don’t know where to start — Change trigger words? Rewrite workflows? Add showcase?
The usual approach is to “polish it for me.” Luban’s approach was to treat it as a work and store it in the workshop: first challenge whether it was worth carving, then see how competitors could stand, measure shortcomings with three rulers, planing each cut through the verification gate, and finally issue the block to set the rules.
What will it deliver?
- A 13-section “Polishing Report”: material inspection conclusions, peer benchmarking table (all with URLs), niche judgment, scoring table, three polishing directions, and directly replaceable rewrite segments
- A screenshot-worthy “Certificate of Mastering”: Scores before and after polishing, a one-sentence new positioning, special skills, next steps
- Validation assets stored in your warehouse: Consolidate one-time comparison scripts into tools, establishing criteria and establishing clear project rules
Five Movements
| Action | What are you doing | A harsh word |
|---|---|---|
| Material Inspection | First, challenge whether the premise of this skill holds | Even rotten wood cannot be carved; if it’s not worth it, just say it directly |
| Visit | Connect online to find peers and see your niche | Behind closed doors you can’t produce good tools |
| Measuring Tapes: | Structure, Measurement, and Living Measurements | Green CI will lie, so you need to reconcile real products |
| Slow Planing | Freeze the baseline, only keep after changing the validation gate | If the quantity doesn’t reach the limit, rehash it; never dig extra just to look like you’ve done too much |
| Reforged | After release, leave a benchmarking watchlist, and in the next round, proceed from feedback | Delivery is not the end point |
Trigger Method
- “Let Luban see this skill” / “Practice it at the classroom”
- “Upgrade my skills” / “Refine my skills”
- “Skill Physical Examination” / “Skill Audit”
- “Why is no one pretending to be my skill?”
- “How to publish this skill to GitHub/ClawHub”
- “Benchmark against similar skills”
Combat Record
This page was translated automatically. Technical details may differ from the original.
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add learnprompt/luban-skill