AgentSkillsCN

Sub Agent

子代理

SKILL.md

Sub-Agent Isolation

Execute tasks in isolated sub-agent context without contaminating the main conversation.

How It Works

This command spawns a dedicated sub-agent to handle your request in complete isolation:

  • •Clean context: Sub-agent starts fresh, no prior conversation history
  • •Parallel execution: Can run multiple sub-agents simultaneously
  • •Result-only return: Only the final output returns to main context
  • •No pollution: Sub-agent's internal processing never affects main conversation

Usage Patterns

Basic Isolation

Simply prefix any request with /sub-agent:

code
/sub-agent analyze this complex codebase for security issues
/sub-agent research best practices for database migrations
/sub-agent generate comprehensive test suite for this module

Parallel Sub-Agents

Run multiple isolated tasks simultaneously:

code
/sub-agent [task1] research authentication patterns
/sub-agent [task2] analyze performance bottlenecks
/sub-agent [task3] generate API documentation

Specialized Sub-Agents

Use specific agent types for better results:

code
/sub-agent --type=research-orchestrator investigate caching strategies
/sub-agent --type=quality-code-analyzer review this module
/sub-agent --type=testing-orchestrator create comprehensive tests

Available Agent Types

Research & Analysis

  • •general-purpose: Complex multi-step tasks
  • •research-orchestrator: Comprehensive research with multi-source validation
  • •quality-code-analyzer: Deep code analysis and pattern detection
  • •infra-context-discovery: Codebase context and pattern mining

Code Generation

  • •generic-coder: Universal coding tasks
  • •typescript-coder: TypeScript with advanced type safety
  • •python-coder: Python with modern best practices
  • •go-coder: Idiomatic Go patterns
  • •rust-coder: Memory-safe Rust development
  • •js-coder: Modern JavaScript/Node.js

Testing & Quality

  • •testing-orchestrator: Comprehensive test orchestration
  • •testing-unit-master: Specialized unit testing
  • •testing-integration-master: Integration test expertise
  • •quality-enforcer: Code quality audit and fixes
  • •quality-security-scan: Security vulnerability detection

Documentation

  • •doc-api-documenter: OpenAPI and endpoint documentation
  • •doc-readme-generator: Professional README creation
  • •doc-module-generator: Module-specific documentation
  • •doc-architecture-designer: Architecture Decision Records (ADRs)

Infrastructure & DevOps

  • •infra-orchestrator: Multi-agent workflow coordination
  • •infra-dependency-manager: Dependency auditing and updates
  • •infra-environment-guardian: Environment consistency checks
  • •cicd-failure-orchestrator: CI/CD failure analysis and fixing

Implementation

When you use /sub-agent, I will:

  1. •Spawn isolated agent with the Task tool
  2. •Execute in clean context without conversation history
  3. •Return only results back to you
  4. •Maintain isolation - no context pollution

Examples

Example 1: Clean Research

code
User: /sub-agent research the latest React 19 features and migration guide