npm package discovery and stats viewer.

Discover Tips

  • General search

    [free text search, go nuts!]

  • Package details

    pkg:[package-name]

  • User packages

    @[username]

Sponsor

Optimize Toolset

I’ve always been into building performant and accessible sites, but lately I’ve been taking it extremely seriously. So much so that I’ve been building a tool to help me optimize and monitor the sites that I build to make sure that I’m making an attempt to offer the best experience to those who visit them. If you’re into performant, accessible and SEO friendly sites, you might like it too! You can check it out at Optimize Toolset.

About

Hi, 👋, I’m Ryan Hefner  and I built this site for me, and you! The goal of this site was to provide an easy way for me to check the stats on my npm packages, both for prioritizing issues and updates, and to give me a little kick in the pants to keep up on stuff.

As I was building it, I realized that I was actually using the tool to build the tool, and figured I might as well put this out there and hopefully others will find it to be a fast and useful way to search and browse npm packages as I have.

If you’re interested in other things I’m working on, follow me on Twitter or check out the open source projects I’ve been publishing on GitHub.

I am also working on a Twitter bot for this site to tweet the most popular, newest, random packages from npm. Please follow that account now and it will start sending out packages soon–ish.

Open Software & Tools

This site wouldn’t be possible without the immense generosity and tireless efforts from the people who make contributions to the world and share their work via open source initiatives. Thank you 🙏

© 2026 – Pkg Stats / Ryan Hefner

evo-anything

v0.1.8

Published

Git-based evolutionary algorithm design engine. Evolves code via LLM-driven mutation, crossover, and reflection on any git repository.

Readme

Evo-anything Plugin — Git-Based Evolutionary Code Optimizer

Evo-anything 是基于论文 "From Understanding to Excelling: Template-Free Algorithm Design through Structural-Functional Co-Evolution"(arXiv:2503.10721)的工程实现。它通过 LLM 驱动的结构-功能协同演化,在任意 git 仓库上自动演化代码,追求更优的 benchmark 表现。

论文引用: Zhe Zhao, Haibin Wen, Pengkun Wang, Ye Wei, Zaixi Zhang, Xi Lin, Fei Liu, Bo An, Hui Xiong, Yang Wang, Qingfu Zhang. From Understanding to Excelling: Template-Free Algorithm Design through Structural-Functional Co-Evolution. arXiv:2503.10721 [cs.SE], 2025.

安装

前置条件

必需:

  • Python >= 3.11
  • Git
  • GitHub CLI (gh) — 用于 /hunt 搜索仓库和自动开 PR

可选(安装后自动启用增强能力):

  • oracle CLI — MapAgent 整仓库上下文分析(npm install -g oracle
  • claude CLI — WorkerAgent 复杂变体生成,用 Claude Code 代替直接 edit
  • codex CLI — WorkerAgent 复杂变体生成的备选
  • lobster CLI — 原子化 setup 工作流 + PR approval gate
  • tmux — 长时间 benchmark 非阻塞后台执行
  • pyflakes — 变体提交前 import/name 静态检查(pip install pyflakes
  • OpenClaw skills: oraclearxiv-watchersummarizesession-logs(通过 clawhub install <slug> 安装)

方式一:npm 一键安装(推荐)

npm install -g evo-anything

安装过程中会自动调用 pip install 完成 Python MCP server 的安装。

安装完成后,运行 setup 配置你的 AI IDE:

# 配置所有支持的平台(Claude Code、Cursor、Windsurf、OpenClaw)
npx evo-anything setup

# 或只配置指定平台
npx evo-anything setup --platform claude
npx evo-anything setup --platform cursor
npx evo-anything setup --platform windsurf
npx evo-anything setup --platform openclaw

方式二:手动安装

通用步骤:安装 evo-engine

无论使用哪个平台,都需要先安装 MCP server:

git clone https://github.com/DataLab-atom/Evo-anything.git
cd Evo-anything/plugin/evo-engine
pip install .

OpenClaw

openclaw plugins install evo-anything
openclaw gateway restart
openclaw plugins doctor   # 验证
openclaw plugins install -l ./plugin
openclaw gateway restart

将插件复制到扩展目录,并在 ~/.openclaw/openclaw.json 中注册:

cp -r plugin/ ~/.openclaw/extensions/evo-anything/
{
  "plugins": {
    "entries": {
      "evo-anything": {
        "enabled": true,
        "config": {}
      }
    }
  },
  "mcpServers": {
    "evo-engine": {
      "command": "evo-engine",
      "args": [],
      "env": {}
    }
  }
}
openclaw gateway restart

验证: 对话中输入 /status,看到 "Evolution not initialized" 即安装成功。


Claude Code

在项目根目录或全局 .claude/settings.json 中添加 MCP server:

{
  "mcpServers": {
    "evo-engine": {
      "command": "evo-engine",
      "type": "stdio"
    }
  }
}

将 skills 链接到 Claude Code:

ln -s $(pwd)/plugin/skills/* ~/.claude/skills/

重启 Claude Code 即可使用。


Cursor

在项目根目录的 .cursor/mcp.json 中添加:

{
  "mcpServers": {
    "evo-engine": {
      "command": "evo-engine",
      "type": "stdio"
    }
  }
}

Cursor 会自动发现 MCP tools(evo_initevo_next_batch 等)。Skills 需要作为 Cursor Rules 手动导入:

cp plugin/AGENTS.md .cursor/rules/evo-agents.md

Windsurf

在全局 ~/.codeium/windsurf/mcp_config.json 中添加:

{
  "mcpServers": {
    "evo-engine": {
      "command": "evo-engine",
      "type": "stdio"
    }
  }
}

其它 MCP 兼容客户端

Evo-anything 的核心是一个标准 MCP server。任何支持 MCP stdio 传输的客户端都可以接入:

# 直接启动 server(stdio 模式)
evo-engine

提供的 MCP tools:evo_initevo_register_targetsevo_report_seedevo_stepevo_next_batchevo_report_fitnessevo_select_survivorsevo_get_statusevo_get_lineageevo_freeze_targetevo_boost_targetevo_record_synergyevo_check_cache


可选配置

演化状态默认存储在 ~/.openclaw/u2e-state/,可通过环境变量自定义(U2E 即论文名 Understanding to Excelling 缩写):

export U2E_STATE_DIR=/path/to/your/state

或在 OpenClaw 中通过 openclaw.json 配置:

{
  "plugins": {
    "entries": {
      "evo-anything": {
        "enabled": true,
        "config": {
          "statePath": "/path/to/your/state"
        }
      }
    }
  }
}

Quick Start

你在 Telegram 发:我要 CIFAR-100-LT 上 SOTA
         ↓
  /hunt 自动触发
         ↓
  搜 GitHub → 找到 3 个候选 → 问你选哪个
         ↓
  你说:用第 1 个
         ↓
  clone → pip install → 下载数据 → 跑基线确认能跑
         ↓
  自动调用 /evolve → 进化循环
         ↓
  每代给你发进度
         ↓
  结束后推最优分支 + 发报告

工作原理

Evo-anything 实现了论文提出的 U2E(Understanding to Excelling)协议——一种无模板的两维协同演化框架,区别于 EoH、FunSearch 等依赖预定义模板、仅做局部函数优化的方法,U2E 同时在功能维(算法逻辑)和结构维(代码架构)上做全局联合优化。

所有实验以 git 分支记录,演化循环包含六个阶段:

  1. 分析 — 自动识别关键算法模块(哪些代码值得优化)
  2. 规划 — 决定变异/交叉策略和每轮变体数量,按温度自适应分配预算
  3. 生成 — LLM 生成代码变体(变异:单亲改进;交叉:双亲融合)
  4. 评估 — 在隔离的 git worktree 中运行 benchmark
  5. 选择 — 保留最优,淘汰其余;每 N 代做跨目标协同(Synergy)检验
  6. 反思 — 提取经验教训,写入结构化记忆,指导后续演化

每一代的最优结果打 tag(best-gen-{N}),最终推送 best-overall 分支。

与现有方法对比

| 方法 | 模板依赖 | 优化范围 | 结构演化 | |------|---------|---------|---------| | EoH / FunSearch | 需要预定义模板 | 局部函数 | 无 | | Evo-anything (U2E) | 无需模板 | 全局多目标 | 功能 + 结构协同 |

Skills

| 命令 | 说明 | |------|------| | /hunt <任务描述> | 搜索 GitHub 找到合适的仓库,自动 clone、安装、跑基线,然后启动演化 | | /evolve <repo> <benchmark_cmd> | 对指定仓库启动演化优化循环 | | /status | 查看当前演化进度 | | /report | 生成完整的演化报告 | | /boost <target_id> | 提升某个优化目标的优先级 | | /freeze <target_id> | 冻结某个目标,停止对它的演化 |

目录结构

Evo-anything/
├── LICENSE
├── README.md
├── research/                  # 生态调研文档
│   ├── 01_openclaw_existing_capabilities.md
│   ├── 02_compatible_products_capabilities.md
│   ├── 03_evo_anything_analysis.md
│   └── 04_ecosystem_capability_map.md  # 生态能力全景图
└── plugin/
    ├── openclaw.plugin.json   # 插件定义
    ├── AGENTS.md              # 演化协议(核心循环)
    ├── SOUL.md                # Agent 人格设定
    ├── TOOLS.md               # 工具使用约定
    ├── agents/                # 各 Agent 行为说明
    │   ├── orchestrator.md    # OrchestratorAgent(含 canvas 可视化)
    │   ├── worker.md          # WorkerAgent(含静态检查、tmux、coding-agent)
    │   ├── policy_agent.md    # PolicyAgent
    │   ├── reflect_agent.md   # ReflectAgent(含跨-run 元学习)
    │   └── map_agent.md       # MapAgent(含 oracle 整仓库分析)
    ├── evo-engine/            # 演化引擎(MCP server)
    │   ├── server.py          # MCP 工具接口
    │   ├── models.py          # 数据模型
    │   └── selection.py       # 选择算法
    ├── skills/                # 用户可调用的技能
    │   ├── hunt/              # 搜索并部署代码库(含 arxiv-watcher)
    │   ├── evolve/            # 启动演化循环(含 lobster 工作流)
    │   ├── status/            # 查看进度
    │   ├── report/            # 生成报告
    │   ├── boost/             # 提升目标优先级
    │   └── freeze/            # 冻结目标
    └── workflows/             # Lobster 声明式工作流
        ├── evo-setup.lobster  # 原子化 setup(validate→baseline→tag→mkdir)
        └── evo-finish.lobster # 结束流程(tag→push→approval gate→PR)

演化记忆

Evo-anything 在目标仓库中维护结构化记忆,避免重复失败的尝试:

memory/
├── global/long_term.md           # 跨目标的通用经验
├── targets/{id}/
│   ├── short_term/gen_{N}.md     # 每代反思
│   ├── long_term.md              # 该目标的累积智慧
│   └── failures.md               # 失败记录(不要再试的方向)
└── synergy/records.md            # 跨函数组合实验结果

分支命名

gen-{N}/{target_id}/{op}-{V}          # 单目标变体
gen-{N}/synergy/{targetA}+{targetB}-{V}  # 跨目标组合

Tags: seed-baseline, best-gen-{N}, best-overall