191341025

191341025 / Self-Evolving-Skill

Public

A design pattern for Claude Code Skills that improve through use — more accurate, more efficient, never bloated. | 越用越准、越用越快、但不臃肿的 Skill 设计模式

13
0
100% credibility
Found Mar 11, 2026 at 12 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

A design pattern and reference setup for AI skills that selectively build and maintain a knowledge base during use, improving efficiency in domains like database exploration without excessive growth.

How It Works

1
📰 Discover the Pattern

You come across a clever idea for making AI helpers smarter as they help you with repeated tasks, like exploring databases.

2
❓ Check If It Fits

You ask yourself if your work involves growing knowledge over time, like learning business rules or data patterns, with a natural limit.

3
📁 Set Up Your Helper

You create a few simple folders: one for instructions, one for tools, and one for a growing notebook of insights.

4
✍️ Write Guiding Rules

You jot down when to use the helper and five simple checks to decide what new knowledge to keep in the notebook.

5
💬 Start Real Tasks

You chat with your AI helper about your actual data questions, like finding patterns in tables or business logic.

6
📈 Watch It Learn

Over sessions, the helper carefully adds useful facts to its notebook only if they pass the checks, staying lean and accurate.

🎉 Smarter Every Time

Your helper now handles familiar tasks quicker and better, remembering key insights without getting cluttered, saving you time.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 12 to 13 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 Self-Evolving-Skill?

Self-Evolving-Skill is a design pattern for Claude Code skills that evolve during use, building a living knowledge base of domain facts like table schemas, query patterns, and business rules. It solves the reset problem in repeated tasks—static skills waste time relearning per session—delivering more accurate results and faster responses without bloating your context window. Built around Claude Code with Python scripts or MCP tools, it includes a MySQL database investigator reference implementation.

Why is it gaining traction?

Unlike traditional static skills or endlessly growing agents, this pattern uses strict governance to add only reusable, non-redundant knowledge, staying lean even as expertise accumulates. Developers dig the selective injection via a compact routing table, which loads just relevant info, mimicking design patterns from the design patterns book for efficient, scalable evolution. The real hook: experiments show 63% rejection rates leading to convergence, making skills mature without manual tweaks.

Who should use this?

Backend engineers probing databases for investigations, like telecom billing audits or smart building analytics. DevOps folks integrating business systems where rules emerge over sessions. Claude Code users in codebase analysis needing persistent structural insights without design github actions system design overhead.

Verdict

Worth prototyping for Claude Code workflows in bounded domains—strong docs, bilingual READMEs, and real experiments prove the concept despite 11 stars and 1.0% credibility score. Still early; lacks broad tests, but the anti-bloated evolution aligns with github.copilot design needs for accurate, stable skills.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.