← Back to Skills Marketplace
Manus Agent
by
zhangifonly
· GitHub ↗
· v1.0.1
· MIT-0
236
Downloads
0
Stars
0
Active Installs
2
Versions
Install in OpenClaw
/install manus-agent
Description
Manus AI Agent 助手，精通通用 AI Agent 任务执行、浏览器操作、代码生成、数据分析
README (SKILL.md)
Manus Agent 助手
你是一个精通 Manus AI Agent 的助手，帮助用户理解和使用 Manus 的自主任务执行能力。
身份与能力
- 精通 Manus 的通用 AI Agent 架构和工作原理
- 熟悉浏览器自动化、代码执行、文件处理等核心能力
- 掌握任务拆解、多步骤执行、结果交付的最佳实践
- 了解 Manus 与 ChatGPT、Claude、Devin 等产品的差异
核心能力
自主任务执行
Manus 能独立完成复杂多步骤任务：
- 自动拆解任务为子步骤
- 在沙盒环境中执行代码
- 操作浏览器完成网页交互
- 生成文件并交付最终结果
浏览器操作
- 自动浏览网页、填写表单、点击按钮
- 抓取和解析网页数据
- 登录网站执行操作（需用户授权）
- 截图记录操作过程
代码生成与执行
- 支持 Python、JavaScript 等主流语言
- 在安全沙盒中运行代码
- 自动安装依赖包
- 处理数据分析、可视化、文件转换等任务
文件处理
- 读取和生成 PDF、Excel、Word、PPT 等格式
- 数据清洗和格式转换
- 图表生成和报告输出
典型使用场景
市场调研
"帮我调研中国新能源汽车市场 Top 10 品牌，整理成 Excel 表格，包含品牌、销量、市占率、主力车型"
数据分析
"分析这份 CSV 销售数据，生成月度趋势图和 Top 产品排名，输出 PDF 报告"
网页开发
"帮我做一个个人作品集网站，包含首页、项目展示、联系方式，风格简约现代"
竞品分析
"对比 Notion、飞书、语雀的功能差异，生成对比表格和分析报告"
与竞品对比
| 维度 | Manus | ChatGPT | Claude | Devin |
|---|---|---|---|---|
| 浏览器操作 | 原生支持 | 插件 | 无 | 有 |
| 代码执行 | 沙盒 | 沙盒 | 无 | 沙盒 |
| 文件交付 | 多格式 | 有限 | 无 | 代码 |
| 多步骤任务 | 优秀 | 一般 | 一般 | 优秀 |
| 中文理解 | 优秀 | 良好 | 优秀 | 一般 |
| 免费额度 | 有限 | 有 | 有 | 无 |
最佳实践
- 任务描述要具体明确，包含预期输出格式
- 复杂任务建议分步骤描述，而非一句话概括
- 涉及网页操作时，提供具体 URL 和操作步骤
- 数据分析任务先上传数据文件，再描述分析需求
- 检查 Agent 的中间步骤，及时纠正方向偏差
- 敏感操作（登录、支付）需要人工确认
最后更新: 2026-03-22
Usage Guidance
This skill is an instruction-only helper and does not install software or ask for secrets by itself, which is good. Before using it: (1) Confirm what runtime tools your agent/platform provides (browser automation, sandboxes, package installs) since the skill assumes those exist. (2) Never paste credentials into a prompt; only provide logins when you intentionally authorize the action and after confirming the exact steps the agent will take. (3) For tasks that install packages or run code, request a preview of the commands the agent will execute and inspect intermediate outputs. (4) If you need stronger guarantees, ask the skill author which concrete tools/APIs the agent will call and whether network requests or file uploads will occur.
Capability Analysis
Type: OpenClaw Skill
Name: manus-agent
Version: 1.0.1
The skill bundle consists solely of metadata and a descriptive markdown file (SKILL.md) that defines a persona for a 'Manus Agent' assistant. It contains no executable code, scripts, or instructions that would trigger malicious behavior, data exfiltration, or unauthorized system access.
Capability Assessment
Purpose & Capability
The skill's name and description claim broad capabilities (browser automation, sandboxed code execution, auto-install of dependencies). As an instruction-only skill it does not request binaries, env vars, or install steps — which is coherent for a 'how to act' helper. However the manifest does not provide or require the tools it describes, so the skill assumes the host agent or platform already exposes browser/code-execution capabilities; that assumption should be verified by the user.
Instruction Scope
SKILL.md instructs the agent to perform multi-step tasks including browser logins and dependency installs and to run code in a sandbox. The instructions are high-level and do not ask the agent to read system files or environment variables. They do allow the agent to request user authorization for logins and to install dependencies — a reasonable runtime behavior but somewhat open-ended. The doc also explicitly recommends manual confirmation for sensitive operations, which mitigates scope creep.
Install Mechanism
No install spec and no code files are present. This is the lowest-risk arrangement: nothing will be written to disk by the skill itself and no external artifacts are fetched.
Credentials
The skill declares no required environment variables, credentials, or config paths. The capabilities listed would normally need user-supplied credentials when performing web logins, but the SKILL.md explicitly states those require user authorization; nothing in the manifest requests unrelated secrets.
Persistence & Privilege
always is false and the skill is user-invocable with normal autonomous invocation allowed. This is the standard default. There is no request for permanent agent presence or modification of other skills/config; no privilege escalation is evident.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install manus-agent - After installation, invoke the skill by name or use
/manus-agent - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.1
- 新增 Manus Agent 技能文档，详细介绍 Manus AI Agent 的核心能力和使用场景
- 覆盖任务自动拆解、浏览器操作、代码生成与执行、文件处理等功能
- 提供典型应用示例：市场调研、数据分析、网页开发、竞品分析等
- 新增 Manus 与 ChatGPT、Claude、Devin 主要竞品的对比表
- 总结最佳实践和提示，帮助用户高效使用 Manus Agent
v1.0.0
- Initial release of "Claude Prompt" skill, version 1.0.0.
- Provides expert guidance for designing high-quality prompts for Claude models.
- Covers XML structuring, system prompt design, chain-of-thought reasoning, and tool use prompts.
- Includes common anti-patterns, advanced tips, and practical best practices.
- Designed to help users in system prompt creation, prompt optimization, and multi-turn dialogue strategies.
Metadata
Frequently Asked Questions
What is Manus Agent?
Manus AI Agent 助手，精通通用 AI Agent 任务执行、浏览器操作、代码生成、数据分析. It is an AI Agent Skill for Claude Code / OpenClaw, with 236 downloads so far.
How do I install Manus Agent?
Run "/install manus-agent" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Manus Agent free?
Yes, Manus Agent is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Manus Agent support?
Manus Agent is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Manus Agent?
It is built and maintained by zhangifonly (@zhangifonly); the current version is v1.0.1.
More Skills