/ai

Usage

`/ai <TASK_DESCRIPTION | list | info <role> | workflow | auto>`

🎯 Task Complexity Assessment

Level 0: Micro Tasks - Direct execution

• **Scenario**: Information queries, file reading, status checks
• **Characteristics**: No code modification, pure information retrieval
• **Boundary**: <5 minutes, no professional knowledge needed
• **Triggers**: "view", "check", "display", "read"
• **Action**: Main controller completes directly, no agent calls

Level 1: Simple Tasks - Single agent direct

• **Scenario**: Single file modification, basic configuration, simple functionality
• **Characteristics**: <50 lines of code, single technology stack, clear requirements
• **Boundary**: One professional domain, no cross-module impact
• **Triggers**: "add", "modify", "configure" single components
• **Action**: Direct call to 1 professional agent, bypassing director

Level 2: Medium Tasks - Single agent complex

• **Scenario**: Complete functional modules, multi-file coordination, requires testing
• **Characteristics**: 50-200 lines of code, requires planning and validation
• **Boundary**: Single technology stack but complex logic, may require refactoring
• **Triggers**: "implement", "develop", "build" complete features
• **Action**: 1 professional agent handles full process, main controller monitors

Level 3: Composite Tasks - Multi-agent serial

• **Scenario**: Cross-module functionality, frontend-backend coordination, 2-3 professional domains
• **Characteristics**: 200-500 lines of code, requires multi-step coordination
• **Boundary**: Clear dependency relationships, serial execution
• **Triggers**: "integrate", "connect", "full-stack" functionality
• **Action**: Main controller serially calls 2-3 agents

Level 4: Parallel Tasks - Multi-agent concurrent

• **Scenario**: Independent module parallel development, performance optimization, multi-platform
• **Characteristics**: 3-5 independent workflows, can execute in parallel
• **Boundary**: Low conflict risk, high independence
• **Triggers**: "simultaneously", "parallel", "multi-platform" development
• **Action**: Main controller calls 3-5 agents in parallel

Level 5: Enterprise Tasks - Director coordination

• **Scenario**: System refactoring, architecture upgrades, complex project analysis
• **Characteristics**: 5+ professional domains, complex dependencies, multi-phase planning
• **Boundary**: Requires specialized task decomposition and coordination management
• **Triggers**: "refactor", "architecture", "system analysis", "enterprise-level"
• **Action**: task-dispatch-director pure coordination, decompose into Level 1-3 tasks

⚡ Auto-Trigger Matrix

**Level 0 Trigger Conditions (no agent calls):**

• Keywords: "view", "check", "display", "read", "list", "status"
• Questions: "what is", "how to understand", "can you explain"
• Operations: Pure information queries, no modification requirements

**Level 1 Trigger Conditions (single agent direct):**

• Keywords: "add", "modify", "update", "configure", "adjust"
• Scope: Single file or component + single technology stack
• Examples: "add Vue component", "modify API endpoint", "configure database connection"

**Level 2 Trigger Conditions (single agent complex):**

• Keywords: "implement", "develop", "build", "create" functional modules
• Scope: Multi-file but single technology stack + requires testing
• Examples: "implement user login", "develop payment module", "build search functionality"

**Level 3 Trigger Conditions (multi-agent serial):**

• Keywords: "integrate", "connect", "full-stack", "end-to-end"
• Scope: 2-3 technology stacks collaboration + clear dependencies
• Examples: "frontend-backend integration", "API integration", "full-stack user system"

**Level 4 Trigger Conditions (multi-agent parallel):**

• Keywords: "simultaneously", "parallel", "multi-platform", "optimize"
• Scope: 3-5 independent modules + low conflict
• Examples: "multi-platform synchronized development", "comprehensive performance optimization"

**Level 5 Trigger Conditions (Director coordination):**

• Keywords: "refactor", "architecture", "system analysis", "enterprise-level", "complete solution"
• Scope: 5+ professional domains + complex planning
• Examples: "system architecture refactoring", "enterprise microservice design", "complex project analysis"

**Mandatory Director Bypass Conditions (Level 0-2):**

• Single file operations
• Clearly specified single technology stack
• User explicitly says "no team collaboration needed"
• Simple information queries and basic modifications

🚫 Direct Handling

Handle without agents:

• File reading, searching, basic analysis
• Simple code modifications or config updates
• Information queries and technical explanations

🛡️ Anti-Over-Engineering Principles

• **One goal, one agent**: Only call one agent unless true collaboration needed
• **Minimum viable solution**: Choose simplest working method
• **User-oriented**: Based on explicit user needs, not assumptions

🎯 Project-Specific Agents Support

Agent Discovery System

The AI system intelligently detects and integrates both:

• **Global Agents**: Standard agents from `/agents/` directory (always available)
• **Project Agents**: Custom agents from `.claude/agents/` directory (created by `/initx`)

Project Agent Features

• **Auto-Detection**: Automatically discovers agents in `.claude/agents/` when present
• **Priority System**: Project-specific agents take precedence over global agents
• **Smart Routing**: Intelligently routes to project agents when they match the task better
• **Seamless Integration**: Works with the same `/ai` command interface

Using Project Agents

# After running /initx to create project-specific agents:
/ai "optimize checkout flow"  # Uses vue-ecommerce-developer if created
/ai "implement payment integration"  # Uses payment-integration-specialist