Tool Security Important

Security mechanisms and permission control for the tool system.

Security Layers

Layer Description

Tool call requests must pass all security layer checks. Rejection at any layer results in call failure.

Risk Levels

Level	Icon	Description	Example
Safe	🟢	Read-only, no side effects	Get time, query info
Medium	🟡	Limited side effects	Send messages
Higher	🟠	Significant actions	Group admin, file ops
Dangerous	🔴	System access	Shell commands

Permission Control

User Permission Levels

Level	Description	Allowed Operations
all	All users	Basic chat, view info
whitelist	Whitelist users	Specific restricted commands
admin	Administrators	Group management, tool calls
master	Bot owner	All operations, system config

Whitelist/Blacklist Control

Dangerous Tools

Dangerous Tool Categories

Dangerous Tool Configuration

builtinTools:
  # Allow dangerous tools
  allowDangerous: false
  
  # Dangerous tool list
  dangerousTools:
    - kick_member
    - recall_message
    - set_group_whole_ban
    - execute_command

Temporary Authorization

Some scenarios require temporary dangerous tool access:

const agent = await createSkillsAgent({
  event: e,
  allowDangerous: true  // Temporary authorization
})

Global Configuration

Enable/Disable Categories

builtinTools:
  enabledCategories:
    - basic
    - user
    - web
  # Unlisted categories are disabled

Disable Specific Tools

builtinTools:
  disabledTools:
    - execute_command
    - delete_file

Preset-Level Control

Whitelist Mode

# Preset file
tools:
  mode: whitelist
  allowedTools:
    - get_time
    - get_weather
    - search_web

Blacklist Mode

tools:
  mode: blacklist
  excludedTools:
    - send_message
    - kick_member

Security Execution Flow

Audit Logging

Log Recording

mcp:
  logging:
    enabled: true
    level: info
    retention: 30  # Keep 30 days

Log Content

{
  "id": "uuid",
  "toolName": "send_message",
  "args": {"target": "123", "content": "..."},
  "result": "success",
  "userId": "456",
  "timestamp": "2024-12-15T06:30:00.000Z",
  "duration": 150
}

View Logs

# Command
#工具日志

# API
GET /api/tools/logs?limit=100&toolName=send_message

Best Practices

Security Recommendations

Follow these best practices to ensure secure tool system operation.

Practice	Description
Least Privilege	Only enable needed tools
Use Whitelist Mode	Use whitelist in presets, explicitly list allowed tools
Disable Dangerous Tools	Always set allowDangerous: false in production
Regular Log Audits	Regularly check tool call logs for anomalies
Strict Parameter Validation	Define parameter Schema to prevent injection attacks

Security Checklist

Pre-deployment Check

Ensure completion of the following security checks:

• [ ] Disabled unnecessary tool categories
• [ ] Dangerous tools disabled or controlled
• [ ] Admin tools properly configured
• [ ] Presets using whitelist mode
• [ ] Audit logging enabled
• [ ] Rate limiting configured

Next Steps

• Built-in Tools - Tool reference
• Custom JS Tools - Create custom tools
• MCP Configuration - MCP settings