How to Run the Development Server for zhangxuefeng-skill: Complete Setup Guide

To run the development server for zhangxuefeng-skill, install an Agent Skills-compatible runtime like Claude Code or Codex, register the skill using the runtime's add command, and execute the dev command to launch a local HTTP server on http://localhost:3000.

The zhangxuefeng-skill repository is not a traditional web application but rather an Agent Skills package designed for AI-agent runtimes. According to the alchaincyf/zhangxuefeng-skill source code, running a development server requires loading this skill into a compatible runtime that understands the Agent Skills protocol, such as Claude Code, Codex, or OpenClaw.

Prerequisites and Runtime Installation

Since zhangxuefeng-skill follows the Agent Skills protocol, you must install a supporting runtime before you can launch a development server. These runtimes expose a local HTTP API that loads and serves skill packages found in their designated directories.

Installing Claude Code (Recommended)

Claude Code provides native support for Agent Skills and is the recommended runtime for development.

npm i -g @agentic/claude-code

Installing Codex

Codex offers an alternative implementation of the Agent Skills protocol.

npm i -g @agentic/codex-cli

Installing OpenClaw

OpenClaw is another compatible runtime that supports the same skill directory structure.

npm i -g @agentic/openclaw

Setting Up the Skill Repository

Before starting the server, clone the repository to your local machine. The skill manifest and supporting assets are located in the root directory.

git clone https://github.com/alchaincyf/zhangxuefeng-skill.git
cd zhangxuefeng-skill

Key files you should know:

• SKILL.md – The skill manifest containing the name, description, and role-play rules
• README.md – High-level overview and installation instructions
• examples/demo-conversation.md – Real-world usage examples
• references/research/ – Source material that the skill references during execution

Running the Development Server

Each runtime provides a dev (or serve) command that starts a local HTTP server. By default, the server binds to http://localhost:3000, though you can override this with the --port flag.

Using Claude Code

First, register the skill with the runtime's global skills directory (typically located at ~/.claude/skills/).

# One-time registration

npx skills add alchaincyf/zhangxuefeng-skill

# Start the development server

claude-code dev

To use a custom port:

claude-code dev --port 4000

Using Codex

Codex uses similar commands but with its own CLI namespace.

# Register the skill

codex add alchaincyf/zhangxuefeng-skill

# Start the server

codex dev

Using OpenClaw

OpenClaw follows the same pattern for skill registration and server startup.

openclaw add alchaincyf/zhangxuefeng-skill
openclaw dev

Verifying the Server is Running

Once the runtime reports the server is ready, verify the skill is properly loaded by sending a test request to the HTTP endpoint.

curl http://localhost:3000/skill/zhangxuefeng-perspective \
     -d '{"prompt":"我想知道该不该学金融"}'

The runtime watches the skill directory for changes, meaning any edits to SKILL.md or files in references/research/ are automatically reloaded without restarting the server. Refresh your browser or resend API requests to see updated behavior immediately.

Summary

• Agent Skills Architecture: The zhangxuefeng-skill package requires a compatible runtime (Claude Code, Codex, or OpenClaw) rather than a standalone server.
• Registration Required: Use the runtime's add command (npx skills add, codex add, or openclaw add) to register the skill before starting the dev server.
• Default Port: Development servers start on http://localhost:3000 by default, configurable via --port.
• Hot Reloading: Changes to SKILL.md and supporting files are automatically detected and reloaded while the server runs.

Frequently Asked Questions

What port does the development server use by default?

The default port is 3000. You can override this by passing the --port flag to the dev command, such as claude-code dev --port 4000 or codex dev --port 8080.

Do I need to restart the server after editing skill files?

No. The runtime implements file watching on the skill directory, so modifications to SKILL.md, references/research/, or other assets are automatically reloaded. Simply refresh your client or resend API requests to see the changes.

Can I use this skill with any AI agent or only specific ones?

The skill requires a runtime that understands the Agent Skills protocol. Compatible options include Claude Code, Codex, Cursor (with Agent Skills extension), OpenClaw, and Hermes Agent. Standard web servers cannot directly load this package.

Where is the skill configuration and metadata stored?

The primary configuration resides in SKILL.md in the repository root. This manifest defines the skill's name, description, and role-play parameters. Additional context is stored in references/research/ and usage examples are provided in examples/demo-conversation.md.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}