Now liveThe Skillselion MCP - thousands of ranked skills, loaded into your agent mid-task. No install.Get it →
miles990 avatar

Evolve

  • 1 installs
  • 1 repo stars
  • Updated January 16, 2026
  • miles990/evolve-plugin

Given a goal, autonomously learns and iterates through a PDCA loop with memory and skill acquisition until the goal is completed.

About

A self-evolving agent skill that takes a goal, runs a PDCA execution loop with git worktree isolation and repo memory, and retries with multiple strategies until success. A developer uses it to hand off open-ended goals like building workflows or raising test coverage.

  • North-star anchoring, PDCA execution, and multi-strategy retry
  • Worktree isolation, repo memory, and skill acquisition modules

Evolve by the numbers

  • 1 all-time installs (skills.sh)
  • Ranked #14,102 of 16,546 AI & Agent Building skills by installs in the Skillselion catalog
  • Data as of Aug 4, 2026 (Skillselion catalog sync)
npx skills add https://github.com/miles990/evolve-plugin --skill evolve

Add your badge

Show developers this skill is listed on Skillselion. Paste this into your README.

Listed on Skillselion
Installs1
repo stars1
Last updatedJanuary 16, 2026
Repositorymiles990/evolve-plugin

What it does

Given a goal, autonomously learns and iterates through a PDCA loop with memory and skill acquisition until the goal is completed.

Files

skill-creator/SKILL.mdMarkdownGitHub ↗

Skill Creator

完整工作流:引導式訪談 → 分析生成 → 驗證 → Token 優化 → 發布到 GitHub

使用方式

/evolve --new-skill "skill 名稱"

五階段流程

Stage 1: 引導式訪談

向使用者提問,收集需求:

1. 問題定義:這個 skill 要解決什麼問題? 2. 目標使用者:新手 / 進階 / 專家? 3. 前置需求:需要什麼 MCP servers 或 CLI tools? 4. 參考來源:有沒有類似的 skill 可以參考?

輸出:內部需求文件

Stage 2: 分析 + 生成

搜尋順序

1. 優先搜尋官方 Skill Repos

2. 如無適合,搜尋 GitHub 上其他 skills 3. 如都無適合參考,使用 4C 方法自行研究

  • 載入 methodology/knowledge-acquisition-4c skill(from claude-domain-skills)
  • Collect:WebSearch / WebFetch 收集官方文檔、最佳實踐
  • Curate:篩選高品質來源,去除噪音
  • Contextualize:分析領域核心概念和常見流程
  • Codify:整理成 skill 可用的知識結構

生成流程

4. 選擇適合的範本(basic / advanced) 5. 生成 SKILL.md 初稿 6. 建立目錄結構(如需要 scripts/templates)

輸出:完整的 skill 目錄

Stage 3: 驗證

檢查清單:

  • [ ] SKILL.md frontmatter 格式正確
  • [ ] 必要欄位存在(name, description, version)
  • [ ] 模擬使用情境,確認指令清楚
  • [ ] 如有 scripts,確認可執行

輸出:驗證報告

Stage 3.5: Token 優化

使用 skill-optimizer 優化新建立的 skill:

1. 分析 token 效率

  • 檢查總行數(目標 < 300 行)
  • 計算核心內容佔比(目標 > 70%)
  • 識別可外連的內容(大型範例、ASCII 圖表、模板)

2. 執行優化

  • 大型 ASCII 圖表 → 簡化為單行描述
  • 完整範例(> 20 行)→ 外連至 extended/examples.md
  • 模板(> 10 行)→ 外連至 extended/templates.md
  • 配置範例 → 外連至擴展檔案

3. 建立分層結構(如需要):

   skill-name/
   ├── SKILL.md           # 核心層 (< 300 行)
   └── extended/          # 擴展層 (按需載入)
       ├── examples.md
       └── templates.md

4. 驗證優化結果

  • 優化前後行數比較
  • 確認功能完整性未受影響

輸出:優化報告(節省 X% tokens)

💡 參考:claude-domain-skills/methodology/skill-optimizer

Stage 4: 發布到 GitHub

1. 詢問:建立新 repo 或加入現有 repo? 2. 生成 README.md 3. git init + commit + push 4. 輸出安裝指令

輸出:

✅ Skill 已發布!

GitHub: https://github.com/<user>/<repo>
安裝: /plugin install <user>/<repo>

範本選擇指南

情況範本
簡單指令、無依賴basic-skill.md
需要 MCP、有複雜流程advanced-skill.md

驗證腳本

./scripts/validate-skill.sh <skill-directory>

發布腳本

./scripts/publish-skill.sh <skill-directory> [--new-repo]

Plugin 格式轉換

將 Skills 倉庫轉換為 Claude Code Plugin Marketplace 格式:

./scripts/convert-to-plugin.sh <skills-repo-path> [--marketplace|--category]

模式:

  • --marketplace:建立 marketplace.json,頂層目錄成為 plugin(推薦)
  • --category:為每個頂層分類建立獨立的 plugin.json

範例:

# 轉換整個 Skills 倉庫為 marketplace
./scripts/convert-to-plugin.sh ~/Workspace/my-skills --marketplace

# 輸出:
# ✅ marketplace.json 已建立
# ✅ 各分類 plugin.json 已建立
#
# 安裝指令:
#   /plugin marketplace add <user>/my-skills
#   /plugin install <plugin-name>@my-skills

Related skills

AI & Agent Buildingagentsautomation

This week in AI coding

Five minutes, every Monday - the tools, releases and tactics for developers.

unsubscribe anytime.