Token导航 LogoToken导航

Agent Skill · 研究检索

smalltalk(smalltalk 搜索)

smalltalk 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

smalltalk 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为接触敏感数据;许可证为 MIT,暂无第三方扫描结果,按安装量在研究检索类的 73,675 个 Skill 里排第 35,283。

总安装 2,112周安装 122GitHub Stars 公开资料未说明下载 783核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
MIT

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:smalltalk(smalltalk 搜索)
来源仓库:https://github.com/clawdbot/skills
仓库路径:skills/smalltalk
安装命令:
npx skills add clawdbot/skills --skill "smalltalk"
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

AgentSkills.tonpx skills
npx skills add clawdbot/skills --skill "smalltalk"

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

用于查找、检索和筛选相关信息。适合在 Codex、Claude、Cursor、Gemini CLI 中快速定位候选结果。适用于通用信息检索和对话场景。可结合来源仓库和原始 README 核验具体用法。通过 npx skills add clawdbot/skills --skill "smalltalk" 安装。

安装前建议确认权限范围和维护状态。

来源仓库内路径:skills/smalltalk

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add clawdbot/skills --skill "smalltalk";命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要接触敏感数据,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

smalltalk 怎么安装?

在终端执行 npx skills add clawdbot/skills --skill "smalltalk"。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

smalltalk 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

smalltalk 需要什么权限?

权限级别是接触敏感数据。该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

smalltalk 是免费的吗?

来源标注的许可证是 MIT。本站不代收任何费用,也不托管安装包;能否商用、要不要署名,以许可证原文为准。

smalltalk 用的人多吗?

公开数据是累计安装 2,112 次,近一周安装 122 次,下载 783 次。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

smalltalk 安全吗?

没有查到第三方扫描结果,安全性无法从公开资料判断,安装前请自行审查源码、依赖和命令行为。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 许可证明确为 MIT,商用和分发边界好判断。

先想清楚的地方

  • 需要接触敏感数据,装到常用环境前要先评估影响面。
  • 没有第三方扫描结果,安全性只能靠自己读源码判断。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 环境里有大量密钥、凭证,不方便给第三方 Skill 读取的人。
  • 要求有安全扫描结论才肯安装的团队。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
smalltalk

smalltalk 搜索 · 本页

2,112—接触敏感数据MIT未收录

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

name
smalltalk
version
1.7.0
description
Interact with live Smalltalk image (Cuis or Squeak). Use for evaluating Smalltalk code, browsing classes, viewing method source, defining classes/methods, querying hierarchy and categories.
metadata
{"clawdbot":{"emoji":"💎","requires":{"bins":["python3","xvfb-run"]}}}

Smalltalk Skill

Execute Smalltalk code and browse live Squeak/Cuis images via MCP.

Prerequisites

Get the ClaudeSmalltalk repo first:

git clone https://github.com/CorporateSmalltalkConsultingLtd/ClaudeSmalltalk.git

This repo contains:

  • MCP server code for Squeak (MCP-Server-Squeak.st)
  • Setup documentation (SQUEAK-SETUP.md, CLAWDBOT-SETUP.md)
  • This Clawdbot skill (clawdbot/)

Setup

  1. Set up Squeak with MCP server — see SQUEAK-SETUP.md
  2. Configure Clawdbot — see CLAWDBOT-SETUP.md

Usage

# Check setup
python3 smalltalk.py --check

# Evaluate code
python3 smalltalk.py evaluate "3 factorial"
python3 smalltalk.py evaluate "Date today"

# Browse a class
python3 smalltalk.py browse OrderedCollection

# View method source (instance side)
python3 smalltalk.py method-source String asUppercase

# View method source (class side)
python3 smalltalk.py method-source "MCPServer class" version
python3 smalltalk.py method-source MCPServer version --class-side

# List classes (with optional prefix filter)
python3 smalltalk.py list-classes Collection

# Get class hierarchy
python3 smalltalk.py hierarchy OrderedCollection

# Get subclasses  
python3 smalltalk.py subclasses Collection

# List all categories
python3 smalltalk.py list-categories

# List classes in a category
python3 smalltalk.py classes-in-category "Collections-Sequenceable"

# Define a new class
python3 smalltalk.py define-class "Object subclass: #Counter instanceVariableNames: 'count' classVariableNames: '' poolDictionaries: '' category: 'MyApp'"

# Define a method
python3 smalltalk.py define-method Counter "increment
    count := (count ifNil: [0]) + 1.
    ^ count"

# Delete a method
python3 smalltalk.py delete-method Counter increment

# Delete a class
python3 smalltalk.py delete-class Counter

Operating Modes

Playground (Default)

Stock image, ephemeral. Changes are discarded when daemon stops. User says: "load Smalltalk skill" or "invoke Smalltalk" — no special flags.

# Start playground daemon
nohup python3 smalltalk-daemon.py start > /tmp/daemon.log 2>&1 &

Dev Mode

User supplies their own image/changes pair. Changes persist across sessions. User says: "load Smalltalk skill in dev mode with ~/MyProject.image"

# Start dev daemon with custom image
nohup python3 smalltalk-daemon.py start --dev --image ~/MyProject.image > /tmp/daemon.log 2>&1 &

Dev mode sets SMALLTALK_DEV_MODE=1 so the MCP server keeps the .changes file (instead of redirecting to /dev/null). The supplied image must have a matching .changes file alongside it.

Common Commands

# Check status
python3 smalltalk.py --daemon-status

# Stop daemon
python3 smalltalk-daemon.py stop

# Restart in dev mode
python3 smalltalk-daemon.py restart --dev --image ~/MyProject.image

Commands

CommandDescription
--checkVerify VM/image paths and dependencies
--daemon-statusCheck if daemon is running
--debugDebug hung system (sends SIGUSR1, captures stack trace)
evaluate <code>Execute Smalltalk code, return result
browse <class>Get class metadata (superclass, ivars, instance methods and classMethods)
method-source <class> <selector> [--class-side]View method source code (supports "Class class" syntax or --class-side flag)
define-class <definition>Create or modify a class
define-method <class> <source>Add or update a method
delete-method <class> <selector>Remove a method
delete-class <class>Remove a class
list-classes [prefix]List classes, optionally filtered
hierarchy <class>Get superclass chain
subclasses <class>Get immediate subclasses
list-categoriesList all system categories
classes-in-category <cat>List classes in a category
explain <code>Explain Smalltalk code (requires ANTHROPIC_API_KEY or OPENAI_API_KEY)
explain-method <class> <sel> [--class-side] [--source <code>]Fetch method from image and explain it (or use --source/--source-file/--source-stdin to bypass daemon)
audit-comment <class> <sel> [--class-side] [--source <code>]Audit method comment vs implementation (or use --source/--source-file/--source-stdin to bypass daemon)
audit-class <class>Audit all methods in a class (instance + class side)
generate-sunit <targets> [--force] [--class-name <name>]Generate SUnit tests for methods and file into image

Environment Variables

VariableDescription
SQUEAK_VM_PATHPath to Squeak/Cuis VM executable
SQUEAK_IMAGE_PATHPath to Smalltalk image with MCP server
ANTHROPIC_API_KEYAPI key for Anthropic Claude (preferred for LLM tools)
ANTHROPIC_MODELAnthropic model (default: claude-opus-4-20250514)
OPENAI_API_KEYAPI key for OpenAI (fallback for LLM tools)
OPENAI_MODELOpenAI model (default: gpt-4o)
LLM_PROVIDERForce LLM provider: anthropic or openai (auto-detected if not set)

Using with Claude Code (MCP mode)

When Claude Code has a live Smalltalk image connected via MCP, explain-method and audit-comment can use pre-fetched source code instead of requiring a running daemon. Use --source, --source-file, or --source-stdin to pass the method source directly:

# Inline source (fetched via MCP, passed on command line)
python3 smalltalk.py explain-method SmallInteger + --source "+ aNumber <primitive: 1> ^ super + aNumber"

# Source from a file
python3 smalltalk.py audit-comment Integer factorial --source-file /tmp/factorial.st

# Source piped via stdin
echo "printString ^ self printStringLimitedTo: 50000" | python3 smalltalk.py explain-method Object printString --source-stdin

The three source flags are mutually exclusive. When none is provided, the daemon is used as before.

Generating SUnit Tests

The generate-sunit command uses an LLM to generate SUnit test cases for Smalltalk methods and files them directly into the running image:

# Generate tests for a single method
python3 smalltalk.py generate-sunit "String>>asUppercase"

# Generate tests for multiple methods
python3 smalltalk.py generate-sunit "Random>>next" "Random>>nextInt:" "Random>>seed:"

# Generate tests for an entire class (all instance methods)
python3 smalltalk.py generate-sunit "OrderedCollection"

# Generate tests for class-side methods
python3 smalltalk.py generate-sunit "Date class>>today"

# Custom test class name
python3 smalltalk.py generate-sunit "String>>asUppercase" --class-name MyStringTests

# Overwrite existing test class
python3 smalltalk.py generate-sunit "String>>asUppercase" --force

# Run the generated tests
python3 smalltalk.py evaluate "StringGeneratedTest buildSuite run printString"

The generated TestCase uses standard SUnit assertions (assert:, assert:equals:, deny:, should:raise:) and is filed into a GeneratedSUnit-* category.

Notes

  • Requires xvfb for headless operation on Linux servers
  • Uses Squeak 6.0 MCP server (GUI stays responsive if display available)
  • saveImage intentionally excluded for safety
  • MCPServer version 7+ required (v7 adds class-side method support)
  • Playground mode: ephemeral, .changes → /dev/null
  • Dev mode: persistent, .changes kept, requires --dev --image PATH

平台分布

windsurf

26.87%
按下载量换算210

Codex

22.15%
按下载量换算173

OpenCode

19.98%
按下载量换算156

Cursor

12.91%
按下载量换算101

Claude Code

8.33%
按下载量换算65

Antigravity

3.77%
按下载量换算30

安全审计

暂无安全审计结果可展示。

权限和风险

接触敏感数据

该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills