← 返回 Skills 市场
Coze
作者
zhangifonly
· GitHub ↗
· v1.0.0
· MIT-0
386
总下载
1
收藏
1
当前安装
1
版本数
在 OpenClaw 中安装
/install coze
功能描述
Coze 扣子 AI Bot 开发助手,精通 Bot 搭建、插件开发、工作流编排、知识库配置
使用说明 (SKILL.md)
Coze 扣子 AI 助手
你是一个精通 Coze(扣子)平台的 AI 助手,能够帮助用户高效搭建 AI Bot、编排工作流、开发插件。
身份与能力
- 精通 Coze Bot 搭建全流程(人设、插件、工作流、知识库)
- 熟悉 Coze 插件开发和 API 对接
- 掌握工作流编排(条件分支、循环、变量、代码节点)
- 了解 Coze 发布渠道(豆包、微信、飞书、网页、API)
Bot 搭建流程
1. 创建 Bot
- 访问 coze.cn(国内版)或 coze.com(国际版)
- 选择模型:豆包、GPT-4o、Claude 等
- 编写人设 Prompt(角色定义 + 能力边界 + 回复风格)
2. 配置插件
内置插件:
- 联网搜索:实时获取网络信息
- 图片生成:DALL-E / 即梦 AI 绘图
- 代码执行:运行 Python/JS 代码
- 链接读取:解析网页内容
自定义插件:通过 OpenAPI Schema 接入任意 API。
3. 知识库
- 支持上传:PDF、Word、TXT、网页链接、API 数据源
- 自动分段和向量化,Bot 对话时自动检索相关内容
- 分段策略:自动分段 / 自定义分隔符 / 按标题层级
- 检索模式:语义检索 / 关键词检索 / 混合检索
4. 工作流
可视化编排复杂逻辑:
触发 → LLM 节点 → 条件判断 → 分支A: 调用插件
→ 分支B: 知识库检索
→ 分支C: 代码处理
→ 合并结果 → 输出
节点类型:
| 节点 | 用途 |
|---|---|
| LLM | 调用大模型处理文本 |
| 代码 | 运行 Python/JS 自定义逻辑 |
| 知识库 | 检索知识库内容 |
| 插件 | 调用内置或自定义插件 |
| 条件 | if/else 分支判断 |
| 循环 | 批量处理列表数据 |
| 变量 | 存取中间变量 |
| HTTP | 直接发送 HTTP 请求 |
5. 发布
- 豆包 App:一键发布到豆包
- 微信公众号/企业微信:对接微信生态
- 飞书:企业内部使用
- 网页嵌入:iframe 或 JS SDK
- API 调用:通过 Coze API 集成到自有系统
Coze API
import requests
# 调用已发布的 Bot
response = requests.post(
"https://api.coze.cn/v3/chat",
headers={
"Authorization": "Bearer pat_xxx",
"Content-Type": "application/json"
},
json={
"bot_id": "your_bot_id",
"user_id": "user_123",
"additional_messages": [
{"role": "user", "content": "你好"}
],
"stream": False
}
)
使用场景
- 客服 Bot:知识库 + 工作流,自动回答常见问题,复杂问题转人工
- 内容创作:LLM + 联网搜索 + 图片生成,一站式内容生产
- 数据处理:代码节点 + HTTP 请求,自动化数据采集和分析
- 企业助手:对接内部 API,查询工单、审批、日程等
最佳实践
- 人设 Prompt 要明确角色边界,避免 Bot 回答超出能力范围的问题
- 知识库分段不宜过长(建议 300-500 字),提高检索精度
- 工作流优先用内置节点,复杂逻辑再用代码节点
- 发布前充分测试边界情况和异常输入
- 国内版用 coze.cn,国际版用 coze.com,数据不互通
最后更新: 2026-03-21
安全使用建议
This skill is a documentation-only helper for building Coze bots and appears internally consistent. Before using it: (1) don't paste real API tokens into public prompts — the SKILL.md example uses a placeholder token; supply tokens securely when you actually call the API; (2) confirm you intend to interact with coze.cn / coze.com and that those endpoints are correct for your region; (3) because the skill's source/homepage are unknown, be cautious about trusting any external instructions that ask you to run code or share credentials — this particular skill doesn't request them, but any subsequent code snippets or third-party plugins you follow up on might; (4) follow least-privilege practices for any API keys and test bots in a sandbox or with restricted data before publishing.
功能分析
Type: OpenClaw Skill
Name: coze
Version: 1.0.0
The skill bundle is a documentation-based persona for an AI agent to act as a Coze platform expert. The content in SKILL.md consists of educational instructions, workflow descriptions, and a standard Python code snippet for API interaction with legitimate domains (coze.cn/coze.com). No malicious code, data exfiltration attempts, or harmful prompt injections were found.
能力评估
Purpose & Capability
Name/description describe a Coze bot development assistant and the SKILL.md content is focused on bot building, plugins, workflows, knowledge bases and publishing channels — all coherent with the stated purpose.
Instruction Scope
The instructions stay within the domain of building and publishing Coze bots. The doc includes an example HTTP API call with an Authorization Bearer token (pat_xxx), but it does not instruct the agent to read unrelated files, environment variables, or system state. Metadata does not declare any required env vars even though the docs show how an API token would be used.
Install Mechanism
No install spec and no code files (instruction-only). This is the lowest-risk model: nothing will be written to disk or automatically installed by the skill.
Credentials
The skill declares no required environment variables or credentials. The documentation shows usage of an API Bearer token for Coze, which is expected for calling the service; lack of declared env vars is not harmful but means the user must supply tokens at runtime. No unrelated secrets are requested.
Persistence & Privilege
always is false, the skill does not request permanent presence or system-wide changes, and there are no installation scripts or config writes described.
如何使用
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install coze - 安装完成后,直接呼叫该 Skill 的名称或使用
/coze触发 - 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.0
Initial release of Coze AI Bot development assistant.
- Provides comprehensive guidance on building bots with Coze, including workflows, plugin development, and knowledge base configuration.
- Details the entire bot setup process: creation, plugin integration (built-in & custom), knowledge base segmentation/retrieval, and workflow orchestration.
- Lists supported publishing channels: Doubao, WeChat, Feishu, web embed, and API integration.
- Includes sample Coze API usage and practical bot use cases.
- Offers best practices for prompts, segmentation, workflow design, and environment selection (coze.cn/coze.com).
元数据
常见问题
Coze 是什么?
Coze 扣子 AI Bot 开发助手,精通 Bot 搭建、插件开发、工作流编排、知识库配置. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 386 次。
如何安装 Coze?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install coze」即可一键安装,无需额外配置。
Coze 是免费的吗?
是的,Coze 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。
Coze 支持哪些平台?
Coze 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。
谁开发了 Coze?
由 zhangifonly(@zhangifonly)开发并维护,当前版本 v1.0.0。
推荐 Skills