requirements-builder（需求构建者）

Requirements Builder

Intelligent requirements gathering system that discovers codebase context, asks structured yes/no questions, and generates comprehensive requirements documentation.

Based on claude-code-requirements-builder.

Arguments

$ARGUMENTS — description of the feature or requirement to gather (e.g., "add user profile photo upload")

Setup

Ensure the requirements/ directory exists in the project root:

mkdir -p requirements
touch requirements/.current-requirement

Workflow Overview

Phase 1: Initial Setup & Codebase Analysis
    → Create timestamped folder, analyze codebase structure
Phase 2: Context Discovery Questions (5 yes/no questions)
    → Ask about problem space, workflows, integrations
Phase 3: Targeted Context Gathering (autonomous)
    → Deep-dive into relevant code, patterns, similar features
Phase 4: Expert Requirements Questions (5 yes/no questions)
    → Ask like a senior dev who knows the codebase
Phase 5: Requirements Documentation
    → Generate comprehensive spec with acceptance criteria

Phase 1: Initial Setup & Codebase Analysis

1. Create timestamp-based folder: requirements/YYYY-MM-DD-HHMM-[slug]
2. Extract slug from $ARGUMENTS (e.g., "add user profile" → "user-profile")
3. Create initial files:

- 00-initial-request.md with the user's request - metadata.json with status tracking

1. Read and update requirements/.current-requirement with folder name
2. Analyze the codebase to understand overall structure:

- Get high-level architecture overview - Identify main components and services - Understand technology stack - Note patterns and conventions

Phase 2: Context Discovery Questions

1. Generate the five most important yes/no questions to understand the problem space:

- Questions informed by codebase structure - Questions about user interactions and workflows - Questions about similar features users currently use - Questions about data/content being worked with - Questions about external integrations or third-party services - Questions about performance or scale expectations - Write all questions to 01-discovery-questions.md with smart defaults - Begin asking questions one at a time proposing the question with a smart default option - Only after all questions are asked, record answers in 02-discovery-answers.md as received and update metadata.json. Not before.

Discovery Question Format:

## Q1: Will users interact with this feature through a visual interface?
**Default if unknown:** Yes (most features have some UI component)

## Q2: Does this feature need to work on mobile devices?
**Default if unknown:** Yes (mobile-first is standard practice)

Phase 3: Targeted Context Gathering (Autonomous)

1. After all discovery questions answered:

- Search for specific files based on discovery answers - Read relevant code in batch - Deep dive into similar features and patterns - Analyze specific implementation details - Use WebSearch and/or context7 for best practices or library documentation - Document findings in 03-context-findings.md including: - Specific files that need modification - Exact patterns to follow - Similar features analyzed in detail - Technical constraints and considerations - Integration points identified

Phase 4: Expert Requirements Questions

1. Now ask questions like a senior developer who knows the codebase:

- Write the top 5 most pressing unanswered detailed yes/no questions to 04-detail-questions.md - Questions should be as if you were speaking to the product manager who knows nothing of the code - These questions are meant to clarify expected system behavior now that you have a deep understanding of the code - Include smart defaults based on codebase patterns - Ask questions one at a time - Only after all questions are asked, record answers in 05-detail-answers.md as received

Expert Question Format:

## Q7: Should we extend the existing UserService at services/UserService.ts?
**Default if unknown:** Yes (maintains architectural consistency)

## Q8: Will this require new database migrations in db/migrations/?
**Default if unknown:** No (based on similar features not requiring schema changes)

Phase 5: Requirements Documentation

1. Generate comprehensive requirements spec in 06-requirements-spec.md:

- Problem statement and solution overview - Functional requirements based on all answers - Technical requirements with specific file paths - Implementation hints and patterns to follow - Acceptance criteria - Assumptions for any unanswered questions

Important Rules

• ONLY yes/no questions with smart defaults
• ONE question at a time
• Write ALL questions to file BEFORE asking any
• Stay focused on requirements (no implementation)
• Use actual file paths and component names in detail phase
• Document WHY each default makes sense

Phase Transitions

• After each phase, announce: "Phase complete. Starting [next phase]..."
• Save all work before moving to next phase

Sub-Commands

/requirements-status

Show current requirement gathering progress and continue from last unanswered question.

1. Read requirements/.current-requirement
2. If no active requirement, suggest starting one
3. If active: show formatted status, load question files, continue from last unanswered question

/requirements-current

Display detailed information about the active requirement (view-only, doesn't continue gathering).

Shows: initial request, codebase overview, all questions/answers, context findings, current phase, next steps.

/requirements-list

Display all requirements with their status and summaries, sorted by active first, then complete, then incomplete.

/requirements-end

Finalize the current requirement gathering session. Options:

1. Generate spec with current information (defaults for unanswered)
2. Mark as incomplete for later
3. Cancel and delete

/requirements-remind

Quick correction when deviating from requirements gathering rules. Auto-detects:

• Open-ended questions asked → rephrase as yes/no
• Multiple questions asked → ask one at a time
• Implementation started → redirect to requirements
• No default provided → add a default

File Structure

Each requirement creates this folder structure:

requirements/
  .current-requirement          # Active requirement folder name
  YYYY-MM-DD-HHMM-[slug]/
    00-initial-request.md       # Original user request
    01-discovery-questions.md   # Context discovery questions
    02-discovery-answers.md     # User's answers
    03-context-findings.md      # AI's codebase analysis
    04-detail-questions.md      # Expert requirements questions
    05-detail-answers.md        # User's detailed answers
    06-requirements-spec.md     # Final requirements document
    metadata.json               # Status tracking

Metadata Structure

{
  "id": "feature-slug",
  "started": "ISO-8601-timestamp",
  "lastUpdated": "ISO-8601-timestamp",
  "status": "active",
  "phase": "discovery|context|detail|complete",
  "progress": {
    "discovery": { "answered": 0, "total": 5 },
    "detail": { "answered": 0, "total": 0 }
  },
  "contextFiles": ["paths/of/files/analyzed"],
  "relatedFeatures": ["similar features found"]
}

Final Spec Format

# Requirements Specification: [Name]

Generated: [timestamp]
Status: [Complete with X assumptions / Partial]

## Overview
[Problem statement and solution summary]

## Detailed Requirements

### Functional Requirements
[Based on answered questions]

### Technical Requirements
- Affected files: [list with paths]
- New components: [if any]
- Database changes: [if any]

### Assumptions
[List any defaults used for unanswered questions]

### Implementation Notes
[Specific guidance for implementation]

### Acceptance Criteria
[Testable criteria for completion]