AgentSkillsCN

clarify

采用结构化的 AskUserQuestion 工作流,深入澄清需求。在实施前,全面收集 MUST_HAVE(关键需求)与 NICE_TO_HAVE(可选需求)信息。当您:(1) 初步启动新功能的实施;(2) 需求表述模糊;(3) 存在多种可行方案;(4) 在尚未编写任何代码之前,均可调用此技能。常用触发词包括:/clarify、“澄清需求”、“提问”、“收集需求”。

SKILL.md
--- frontmatter
# VERSION: 2.43.0
name: clarify
description: "Intensive requirement clarification using structured AskUserQuestion workflow. Gathers MUST_HAVE (blocking) and NICE_TO_HAVE (optional) information before implementation. Use when: (1) starting new feature implementation, (2) requirements are ambiguous, (3) multiple approaches possible, (4) before writing any code. Triggers: /clarify, 'clarify requirements', 'ask questions', 'gather requirements'."
user-invocable: true

Clarify - Intensive Questioning (v2.37)

Systematically gather requirements using TLDR semantic search + AskUserQuestion tool.

Quick Start

bash
/clarify  # Start intensive questioning for current task

Pre-Clarification: TLDR Semantic Search (v2.37)

AUTOMATIC - Before asking questions, use semantic search to understand existing code:

bash
# Find existing related functionality (95% token savings)
tldr semantic "$USER_TASK_KEYWORDS" .

# Example: For "add authentication", find existing auth code
tldr semantic "authentication login session user password" .

# Get structure overview for context
tldr structure . --lang "$PRIMARY_LANGUAGE"

This helps formulate better questions based on what already exists in the codebase.

Workflow

MUST_HAVE Questions (Blocking)

These MUST be answered before proceeding:

yaml
AskUserQuestion:
  questions:
    - question: "What is the primary goal of this feature?"
      header: "Goal"
      multiSelect: false
      options:
        - label: "New user-facing feature"
        - label: "Internal refactoring"
        - label: "Bug fix"
        - label: "Performance optimization"

Categories to Cover

  1. Functional Requirements

    • What exactly should this do?
    • What are inputs/outputs?
    • Edge cases?
  2. Technical Constraints

    • Existing patterns to follow?
    • Technology preferences?
    • Performance requirements?
  3. Integration Points

    • Existing code interactions?
    • APIs to maintain?
    • Database changes?
  4. Testing & Validation

    • How will this be tested?
    • Acceptance criteria?
  5. Deployment

    • Feature flags needed?
    • Rollback strategy?

NICE_TO_HAVE Questions

Accept defaults but still ask:

yaml
AskUserQuestion:
  questions:
    - question: "Implementation preferences?"
      header: "Approach"
      multiSelect: true
      options:
        - label: "Minimal changes"
        - label: "Include tests"
        - label: "Add documentation"

Question Templates

Goal Clarification

yaml
AskUserQuestion:
  questions:
    - question: "What problem does this solve?"
      header: "Problem"
      options:
        - label: "User pain point"
          description: "Direct user-facing issue"
        - label: "Technical debt"
          description: "Code maintainability"
        - label: "Performance issue"
          description: "Speed/resource usage"
        - label: "Security concern"
          description: "Vulnerability fix"

Scope Definition

yaml
AskUserQuestion:
  questions:
    - question: "What is the scope?"
      header: "Scope"
      options:
        - label: "Single file"
        - label: "Single module"
        - label: "Multiple modules"
        - label: "Cross-system"

Priority

yaml
AskUserQuestion:
  questions:
    - question: "Priority level?"
      header: "Priority"
      options:
        - label: "Critical (blocking)"
        - label: "High (this sprint)"
        - label: "Medium (this quarter)"
        - label: "Low (backlog)"

Integration

  • Invoked by /orchestrator in Step 1
  • Pre-step: tldr semantic search (automatic in v2.37)
  • Must complete before CLASSIFY step
  • Results inform plan complexity

TLDR Integration (v2.37)

PhaseTLDR CommandPurpose
Before questionstldr semantic "$KEYWORDS" .Find related code
Context gatheringtldr structure .Codebase overview
Dependency checktldr deps "$FILE" .Impact analysis

Anti-Patterns

  • Never proceed with unanswered MUST_HAVE questions
  • Never assume user intent
  • Never skip clarification for features
  • Never ask more than 4 questions at once (AskUserQuestion limit)