Custom Agent

更新时间:
复制 MD 格式

Custom agents extend base capabilities by handling specific tasks with isolated context windows, tool permissions, and system prompts. They are managed as subagents.

Create a Custom Agent

Method 1: Use create-agent (recommended)

Use the built-in create-agent skill to interactively create a custom agent:

/create-agent <your goal, for example, code review expert>

The create-agent skill guides you through these steps:

  • Define the agent name and description

  • Select required tool permissions

  • Auto-generate a system prompt template

  • Save the agent file to the correct location

Note

For your first custom agent, use /create-agent to generate a correctly formatted configuration file.

Method 2: Create manually

Alternatively, create a .md file manually in one of these locations:

Location

Path

Scope

User-level

~/.lingma/agents/<agentName>.md

All projects

Project-level

${project}/.lingma/agents/<agentName>.md

Current project only

Each agent file requires a frontmatter block for metadata and a system prompt:

---
name: code-review
description: A code review expert that checks code quality and security
tools: Read, Grep, Glob, Bash
---
You are a senior code reviewer. Ensure high code quality. Review checklist:
1. Code readability
2. Naming conventions
3. Error handling
4. Security checks
5. Test coverage

Field

Required

Description

name

Yes

Unique agent identifier

description

Yes

Brief description of function and expertise. Used for automatic agent selection

tools

No

List of allowed tools, comma-separated

Supported Tools

Tool Name

Description

Shell

Run shell commands

Edit

Edit files

Write

Create or overwrite files

Glob

Find files by pattern

Grep

Search file contents

Read

Read files

WebFetch

Fetch URL content

WebSearch

Search the web with domain filtering

Use in Your IDE

Method 1: Automatic Trigger

Describe your task in natural language. The model matches your intent to the appropriate agent:

Review the implementation of this API

The model detects your intent and invokes the code-review agent.

Method 2: Manual Trigger

Type /agent-name to trigger a specific agent:

/code-review