prompt-analyzer

M2AI-jp's avatarfrom M2AI-jp

ユーザープロンプトの深層分析を行う専門Skill。5W1H抽出、リスク分析、曖昧さ検出を実施し、pm SubAgentに構造化データを返す。

⭐1stars🔀2forks📁View on GitHub🕐Updated Jan 1, 2026⬇️1downloads

When & Why to Use This Skill

The Prompt Analyzer is a specialized Claude skill designed for deep semantic analysis of user inputs to optimize agentic workflows. It transforms raw user prompts into structured data by performing 5W1H extraction, risk assessment, and ambiguity detection. By accurately classifying intents into instructions, questions, or context, it enables precise task routing and ensures that sub-agents or playbooks receive high-quality, actionable information.

Use Cases

  • Intent Classification & Routing: Automatically categorize user prompts to determine whether to trigger a specific playbook, provide a direct answer, or integrate information into the current context.
  • Requirement Clarification: Identify missing 5W1H details and ambiguities in complex user requests to prevent execution errors and improve output accuracy.
  • Risk & Blocking Issue Detection: Analyze incoming instructions for potential risks or logical gaps before they reach the execution phase, acting as a pre-processing guardrail.
  • Multi-Instruction Decomposition: Break down complex, multi-faceted user prompts into individual logical points for systematic processing by specialized sub-agents.
nameprompt-analyzer
descriptionユーザープロンプトを分析し、topic_type(instruction/question/context)を判定する。

Prompt Analyzer Skill

ユーザープロンプトを分析し、構造化データを返す Skill。

Purpose

  • topic_type 判定: プロンプトを instruction / question / context に分類
  • 5W1H + リスク + 曖昧さ検出: タスク依頼の詳細分析
  • 論点分解: 複数指示の分離
  • pm 支援: playbook 作成前の情報収集

When to Use

triggers:
  - 全てのユーザープロンプト(Hook から強制呼び出し)

invocation:
  Task(subagent_type='prompt-analyzer', prompt='ユーザープロンプト')

Output

analysis:
  summary:
    primary_topic_type: instruction|question|context
    confidence: high|medium|low
    ready_for_playbook: true|false
    blocking_issues: ["{問題}"]
  必須アクション:
    - id: 1
      action: "この分析結果をチャットに出力"
    - id: 2
      action: "{topic_type に基づくアクション}"

Next Action

instruction:
  action: Skill(skill='playbook-init')
question:
  action: 直接回答
context:
  action: 現在タスクに統合
ファイル 役割
agents/prompt-analyzer.md SubAgent 詳細定義
.claude/hooks/prompt.sh Hook(呼び出し元)