AgentSkillsCN

task

智能任务路由系统，可自动选择最优代理并执行任务。通过自然语言解析任务描述，精准匹配项目上下文，为每项任务推荐最合适的代理或执行命令。 【适用场景】当用户提及“task”“任务执行”，或在提供自然语言任务描述时，又或是需要为开发任务配备智能代理路由功能时，可运用此技能。 【关键词】task、智能路由、代理选择、自然语言、任务执行、编排

SKILL.md
--- frontmatter
name: task
description: |
  Intelligent task router with automatic agent selection. Analyze natural language task descriptions and route to optimal agent/command.
  [What] Parse task intent, integrate project context, select optimal agent, and execute with appropriate strategy
  [When] Use when: users mention "task", "タスク実行", provide natural language task descriptions, or need intelligent agent routing for development tasks
  [Keywords] task, intelligent routing, agent selection, natural language, task execution, orchestration

Task - Intelligent Task Router

Overview

Analyze natural language task descriptions and automatically route to the optimal agent/command for execution.

Core Mission

Parse natural language tasks, understand project context, and select the optimal agent/command/toolchain to deliver practical results.

Processing Architecture

Phase 1: Multi-Layer Task Analysis

Analyze tasks from multiple perspectives:

Intent Analysis

  • •Primary intent detection (implementation, debugging, review, etc.)
  • •Task category classification
  • •Complexity assessment

Structural Decomposition

  • •Identify targets, constraints, scope
  • •Analyze dependencies
  • •Determine if simple or complex task

Phase 2: Dynamic Context Integration

Integrate project information and execution history:

  • •Auto-detect project type (Next.js, React, Go, etc.)
  • •Load technology stack information
  • •Enhance with execution history
  • •Validate constraints

Phase 3: Intelligent Agent Selection

Select optimal agent with confidence-based scoring:

Simple Tasks (complexity < 0.8):

  • •Single agent execution
  • •Direct task completion

Complex Tasks (complexity ≥ 0.8):

  • •Multi-agent coordination
  • •Task decomposition and parallel execution

Agent Capability Mapping:

  • •error-fixer: Error fixing, type safety, code quality
  • •orchestrator: Implementation, refactoring, task decomposition
  • •code-reviewer: Code review, quality assessment, security
  • •researcher: Investigation, analysis, root cause analysis
  • •docs-manager: Documentation management, link validation
  • •serena: Semantic analysis, symbol search

Phase 4: Execution & Optimization

Execute with real-time optimization:

  • •Display execution plan
  • •Execute single or multi-agent strategy
  • •Enhance results with context
  • •Record metrics (execution time, quality score, etc.)
  • •Save context for future reference

Advanced Features

Deep Thinking Mode

Enable with --deep-think or --thinking flags:

  • •Enhanced analysis for complex tasks
  • •Focus areas: root cause analysis, design decisions, optimization strategies, implementation strategies
  • •Complexity-based threshold (0.7)

Focus Area Detection:

  • •Root cause: "なぜ", "why", "原因", "cause"
  • •Design: "設計", "design", "アーキテクチャ", "architecture"
  • •Optimization: "最適", "optimal", "改善", "improve"
  • •Implementation: "実装", "implement", "方法", "method"

Continuous Learning System

Record execution results and improve future accuracy:

  • •Track task patterns and success rates
  • •Record agent performance metrics
  • •Generate recommendations based on history
  • •Calculate expected time and best agent

Usage Examples

Basic Usage

bash
# Natural language task specification
/task "このコードをレビューして品質を確認"
/task "ユーザー認証機能を実装"
/task "パフォーマンスを改善"

# Git/branch-related reviews
/task "origin/developでレビューして"
/task "最新のコミットをレビュー"

Advanced Usage

bash
# Multi-step tasks
/task "新機能を実装してテストを書いてドキュメントも更新"

# Constrained tasks
/task "Go言語でClean Architectureに従ってREST APIを実装"

# Analysis tasks
/task "なぜこのテストが失敗するのか原因を調査して修正案を提示"

# Semantic analysis tasks
/task "AuthServiceインターフェースの全ての実装を見つけて"
/task "getUserByIdメソッドを呼び出している全ての場所を探して"

Interactive Mode

bash
# Interactive execution
/task --interactive "複雑な問題を解決"

# Dry run
/task --dry-run "大規模リファクタリング"

# Verbose logging
/task --verbose "パフォーマンス最適化"

# Deep Thinking mode
/task --deep-think "複雑な技術判断が必要なタスク"
/task --thinking "なぜこのエラーが発生するか調査"

Integration Points

Shared Utilities

This command integrates with shared utilities:

  • •shared/task-context.md: Unified task context
  • •shared/agent-selector.md: Agent selection logic
  • •shared/project-detector.md: Project type detection

Skill Integration

Auto-loads relevant skills based on task analysis:

Framework Skills:

  • •integration-framework: TaskContext standardization, Communication Bus patterns

Technology Stack Skills (auto-detected):

  • •typescript: Type safety, any-type elimination, Result<T,E> patterns
  • •react: Component design, Hooks, performance optimization
  • •golang: Idiomatic Go, error handling, concurrency
  • •security: OWASP Top 10, input validation, authentication

Execution Flow

code
/task "TypeScript型エラーを修正"
    ↓
TaskContext creation (project detection)
    ↓
Technology stack detection: TypeScript
    ↓
Auto-load skills: ["typescript", "code-quality-improvement"]
    ↓
Agent selection: error-fixer
    ↓ (with skill context)
TypeScript type safety patterns + 3-layer fix strategy
    ↓
Execution complete

Notes

  • •Always executes via agent-based execution (no direct command execution)
  • •All output is in Japanese
  • •Metrics tracked: execution time, quality score, resource usage
  • •Learning system records all executions for future improvement