DazhuangJammy

基于 Claude Code 官方 skill-creator,我融入了自己对提示词架构、Skill 架构以及 CLI 工具运行机制的理解,对其进行了全面重构,打造了一个新版 Skill Creator。 在测评环节,我采用 Codex 的 Headless 模式(无需打开图形界面,直接在终端执行)进行了三轮以上的独立对话测试。测试结果显示,新版在各项维度上全面超越原版。完整的测试标准与评测报告已归档至「测评报告」文件夹,可供参考。 此外,新版在可维护性上也有显著提升——原版 Skill 一旦生成,提示词后续极难修改和维护;而新版的架构设计使得迭代和调整更加便捷,也更利于团队协作和长期维护。欢迎 star,作者 vx:yinyinGyL

29
3
100% credibility
Found Apr 10, 2026 at 29 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
HTML
AI Summary

A collection of helper scripts for creating structured AI skill blueprints, testing their behavior, optimizing descriptions, and packaging them for use in Claude-based chat tools.

How It Works

1
🔍 Discover the skill maker

You find this helpful tool on a sharing site that makes creating AI assistants easier and better.

2
📥 Add it to your workspace

You bring the tool into your AI helper folder so it's ready to use alongside your other assistants.

3
🆕 Start a new assistant blueprint

You create a ready-made outline for your new AI assistant, picking what extra helpers it needs.

4
✏️ Fill in your ideas

You add simple instructions on what problems your assistant solves, like a clear recipe for everyday tasks.

5
Quickly check it works

You run a fast review to make sure your assistant follows the right structure and rules.

6
🎯 Test and improve triggering

You try sample questions to see if your AI picks your assistant at the perfect times, then tweak the summary until it shines.

🎉 Package and share your assistant

Your polished AI assistant is bundled up, ready to use, share with friends, or add to your collection.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 29 to 29 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is DazhuangSkill-Creator?

DazhuangSkill-Creator is a Python CLI toolkit for building Claude Code skills, rebuilt from the official skill-creator with better prompt and skill architecture for easier iteration and maintenance. It scaffolds skills via commands like init_skill, validates structure, runs headless benchmarks against Codex, and optimizes descriptions through eval loops to improve triggering accuracy. Users get packaged .skill files ready for Claude Code install, plus auto-update checks for staying current.

Why is it gaining traction?

Benchmarks show it outperforming the official version (99.4 vs 96.2 score) across content creation, structured output, tool research, script execution, and orchestration, with smaller skill sizes and cross-platform Windows/macOS support. The optimization loop uses train/test splits to refine descriptions without overfitting, and a live HTML viewer tracks progress. Developers hook on the CLI-driven workflow for claude code skills that evolve with teams, unlike rigid originals.

Who should use this?

Claude Code users crafting custom skills for GitHub integration, code review, or CLI automation, especially those tired of brittle prompts that derail in long contexts. Backend devs building claude github actions or connectors will like the script execution focus; AI prompt engineers benefit from eval tools for precise triggering.

Verdict

Grab it if you're deep into Claude Code skills—low 29 stars and 1.0% credibility reflect early maturity, but solid docs, benchmarks, and Apache 2.0 license make it worth a test drive over the official. Skip for production without your own evals.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.