Zhin Tool Service Guide

Use this skill to create, register, and manage tools in Zhin plugins. The Tool Service provides a unified abstraction that bridges AI agent tool-calling and traditional message commands.

Core Concepts

A Tool in Zhin can serve three purposes simultaneously:

1. AI Agent tool — callable by LLM agents via function calling
2. Message Command — executable by users as a chat command
3. MCP tool — exposable via the Model Context Protocol

ZhinTool Class (Chain API)

ZhinTool provides a builder-style API similar to MessageCommand:

import { usePlugin, ZhinTool } from 'zhin.js'

const plugin = usePlugin()

const weatherTool = new ZhinTool('weather')
  .desc('Query weather information')
  .param('city', { type: 'string', description: 'City name' }, true)
  .param('days', { type: 'number', description: 'Forecast days' })
  .platform('qq', 'telegram')
  .scope('group', 'private')
  .permission('user')
  .usage('Query the weather for a city')
  .examples('weather Beijing', 'weather Tokyo 3')
  .alias('天气')
  .execute(async (args, ctx) => {
    return `Weather in ${args.city}: Sunny`
  })
  .action(async (message, result) => {
    return `Weather: ${result.params.city}`
  })

plugin.addTool(weatherTool)

ZhinTool Methods

Method	Description
.desc(text)	Set tool description (for AI and help)
.param(name, schema, required?)	Add a parameter (ordered)
.platform(...names)	Restrict to platforms (e.g. 'qq', 'telegram')
.scope(...types)	Restrict to scene types ('private', 'group', 'channel')
.permission(level)	Set minimum permission level
.permit(...perms)	Add legacy permission strings (compatible with MessageCommand)
.tag(...tags)	Add tags for filtering
.hidden(value?)	Hide from help listings
.usage(...texts)	Add usage descriptions
.examples(...texts)	Add usage examples
.alias(...names)	Add command aliases
.pattern(pat)	Override the auto-generated command pattern
.execute(fn)	Required. Set the AI tool execution function (args, ctx) => result
.action(fn)	Set the command callback (message, matchResult) => result

execute vs action

• execute (required) — Called when the tool is invoked by an AI agent. Receives (args, context?) where args are parsed parameters and context is a ToolContext.
• action (optional) — Called when the tool is invoked as a chat command. Receives (message, matchResult) just like MessageCommand.action. If omitted, execute is used as a fallback to auto-generate the command handler.

defineTool Helper

Use defineTool for a type-safe object-style definition:

import { usePlugin, defineTool } from 'zhin.js'

const plugin = usePlugin()

const calcTool = defineTool<{ expression: string }>({
  name: 'calculator',
  description: 'Calculate a math expression',
  parameters: {
    type: 'object',
    properties: {
      expression: {
        type: 'string',
        description: 'Math expression',
      },
    },
    required: ['expression'],
  },
  command: {
    pattern: 'calc <expression:rest>',
    alias: ['calculate'],
    usage: ['Calculate math expressions'],
  },
  execute: async (args) => {
    // args.expression has full type inference
    const math = await import('mathjs')
    return `Result: ${math.evaluate(args.expression)}`
  },
})

plugin.addTool(calcTool)

The generic parameter <{ expression: string }> gives you type checking on args inside execute.

Registering Tools

addTool — register with command generation

plugin.addTool(tool)

Registers the tool and auto-generates a corresponding MessageCommand (unless command: false).

addToolOnly — register without command

plugin.addToolOnly(tool)

Registers the tool for AI agent use only, without generating a command.

Both methods return a dispose function:

const dispose = plugin.addTool(weatherTool)
// Later: dispose() removes the tool and its command

Tools are automatically cleaned up when the plugin is disposed.

Tool ↔ Command Conversion

The ToolService can convert in both directions:

Tool → Command (automatic)

When you call plugin.addTool(tool), a MessageCommand is auto-generated from the tool's parameters:

// Tool with parameters: { city: string (required), days: number (optional) }
// Generates command pattern: "weather <city:text> [days:number]"

Parameter ordering: required parameters come first, then optional ones.

Command → Tool (via ToolService)

Existing commands can be wrapped as tools for AI agent use:

const toolService = plugin.inject('tool')
const tool = toolService.commandToTool(existingCommand, 'my-plugin')

collectAll — gather all available tools

const toolService = plugin.inject('tool')
const allTools = toolService.collectAll(plugin)

This collects tools from three sources:

1. ToolService tools — tools registered via addTool/addToolOnly
2. Commands — existing MessageCommand instances converted to tools
3. Adapter tools — platform-specific tools from adapters

Permission Levels

Tools support a hierarchical permission system (from lowest to highest):

Level	Priority	Description
user	0	Regular user (default)
group_admin	1	Group administrator
group_owner	2	Group owner
bot_admin	3	Bot administrator
owner	4	Zhin instance owner

new ZhinTool('admin-action')
  .desc('Admin-only operation')
  .permission('bot_admin')
  .execute(async (args) => { /* ... */ })

The ToolService automatically checks permission levels when filtering tools for a given context.

Platform and Scope Filtering

Restrict where a tool can be used:

new ZhinTool('group-only')
  .platform('qq', 'discord')       // Only on QQ and Discord
  .scope('group')                   // Only in group chats
  .execute(async (args) => { /* ... */ })

The filterByContext method checks all restrictions:

const toolService = plugin.inject('tool')
const context = {
  platform: 'qq',
  scope: 'group',
  senderPermissionLevel: 'user',
}
const available = toolService.filterByContext(allTools, context)

ToolContext

When a tool is executed, it receives a ToolContext with runtime information:

interface ToolContext {
  platform?: string        // Source platform (e.g. 'qq')
  botId?: string           // Bot identifier
  sceneId?: string         // Scene ID (group/channel/user)
  senderId?: string        // Sender's user ID
  scope?: ToolScope        // 'private' | 'group' | 'channel'
  message?: Message        // Original message object
  senderPermissionLevel?: ToolPermissionLevel
  isGroupAdmin?: boolean
  isGroupOwner?: boolean
  isBotAdmin?: boolean
  isOwner?: boolean
  extra?: Record<string, any>
}

ToolService API

Access the ToolService via context injection:

const toolService = plugin.inject('tool')

Method	Description
add(tool, pluginName, generateCommand?)	Register a tool
remove(name)	Remove a tool by name
get(name)	Get a tool by name
getAll()	Get all registered tools
getByTags(tags)	Filter tools by tags
execute(name, args, context?)	Execute a tool by name
commandToTool(command, pluginName)	Convert a Command to a Tool
collectAll(plugin)	Collect all tools from all sources
filterByContext(tools, context)	Filter tools by platform/scope/permission

Parameter Schema

Tool parameters use JSON Schema format with Zhin extensions:

{
  type: 'object',
  properties: {
    city: {
      type: 'string',
      description: 'City name',
      paramType: 'text',    // Zhin extension: command param type
    },
    count: {
      type: 'number',
      description: 'Number of results',
      default: 5,
    },
    format: {
      type: 'string',
      enum: ['json', 'text'],
      description: 'Output format',
    },
  },
  required: ['city'],
}

The paramType extension controls how the parameter is parsed in command mode:

• 'text' (default for strings)
• 'number' (default for numbers)
• 'boolean'
• 'rest' (captures all remaining text)

Serialization

toJSON — for AI function calling

const tool = new ZhinTool('weather').desc('...').param(...)
const schema = tool.toJSON()
// Returns: { name, description, parameters, platforms?, scopes?, permissionLevel?, tags? }
// Compatible with OpenAI Function Calling format

help — human-readable help text

console.log(tool.help)
// weather <city:text> [days:number]
//   Query weather information
//   Parameters:
//     city: string (required) City name
//     days: number (optional) Forecast days

Disabling Command Generation

Set command: false to prevent auto-generating a command:

const tool = defineTool({
  name: 'internal_lookup',
  description: 'Internal data lookup',
  parameters: { type: 'object', properties: {} },
  command: false,
  execute: async () => { /* AI-only tool */ },
})

Checklist

• Use ZhinTool (chain API) or defineTool (object API) to create tools.
• Always provide execute() — it is required for AI agent invocation.
• Use action() on ZhinTool for custom command behavior; omit it to auto-generate from execute.
• Register with plugin.addTool() (tool + command) or plugin.addToolOnly() (tool only).
• Set permission(), platform(), and scope() to control access.
• Use collectAll() and filterByContext() when building AI agent tool lists.
• Tools are auto-disposed when their parent plugin is disposed.