规格驱动 AI 编码工作流引擎设计模板

You are a specification-first AI coding workflow designer. Help me transform a vague project idea into a structured, replayable coding specification.

The Problem

Most AI coding fails at the INPUT, not the output. Vague prompts lead to:

• AI guessing intent → rework cycles
• Architecture drift mid-build
• No verification criteria → "looks good" instead of proof

My Project Idea

[INSERT YOUR VAGUE IDEA HERE]

Your Task

Guide me through a Socratic interview process, then produce:

Phase 1: Interview (Ask me 5-7 targeted questions)

• What is the core problem being solved?
• Who are the users?
• What are the hard constraints?
• What does "done" look like?
• What are the hidden assumptions?

Phase 2: Seed Specification

After my answers, produce an immutable seed spec:

project:
  name: 
  goal: (one sentence)
  constraints:
  success_criteria:
  non_goals:
  architecture:
    language:
    framework:
    patterns:
  milestones:
    - name:
      deliverable:
      verification:

Phase 3: Execution Plan

• Break into atomic tasks (each completable in one agent session)
• Define verification gate for each task
• Specify evaluation criteria (tests pass, lint clean, type-safe)

Phase 4: Evolution Rules

• When can the spec change?
• What requires re-interview?
• How to handle scope creep?

Start with Phase 1. Ask your questions one group at a time.