yjj0333/skills-manager-backup/ai-project-briefing

AI Project Briefing

Guide the user through project initiation before any code is written. Discussion happens in Chinese; output documents and AI rules are generated in English.

Core principle: Do not start coding from a one-sentence idea. First turn the idea into a clear project brief, MVP boundary, user flow, data object list, and acceptance criteria.

Pipeline Position

This is Stage 1 of 9 in the AI Project Toolkit pipeline:

1. ai-project-briefing — clarify product idea, MVP, scope, flows, business objects
2. ai-tech-advisor — choose the technical route and stack
3. ai-db-designer — design database from business objects and flows
4. ai-frontend-scaffolder — design frontend skeleton and UI rules
5. ai-backend-api-planner — design backend responsibilities, API boundaries, auth, validation
6. backend-skeleton-builder — build minimal runnable backend skeleton with rules-first approach
7. backend-architecture-reviewer — verify and accept the backend architecture
8. backend-security-checkpoint — audit API and permission security
9. backend-deploy-checkpoint — pre-launch checklist: secrets, build & start, database backup + migration, smoke test, rollback

If later stage specs exist, read them as context and offer to update the brief. If this stage has no specs/project-brief-spec.md, create it before recommending code work.

Auto-Detect Existing Context

Before discussion, scan the project root and specs/ for (older projects may keep these at the project root):

Interaction Flow

1. Create/open project space — ensure the user has a named project folder
2. Clarify product idea — what it is, who uses it, what problem it solves
3. Explore user scenarios — key users, jobs-to-be-done, main workflows
4. Define MVP boundary — first version must-have, later, explicitly not now
5. Identify business objects — users, orders, tasks, articles, files, permissions, etc.
6. Sketch user flow and data flow — core path from entry to completion
7. Explore differentiation — similar products, opportunities, value proposition
8. Set milestones and acceptance criteria — what counts as first version done
9. Generate outputs — project brief, AI rules, prompt templates
10. Offer next stage — recommend ai-tech-advisor

Adaptive Depth Strategy

• If the user gives a clear product description, summarize and move quickly.
• If the idea is vague, ask one question at a time.
• If the user wants to code immediately, pause and explain that this stage is for discussion only.
• If the project is too large, decompose it into sub-projects and brief the first one.

Required First Question

Ask:

你的项目一句话是什么？它面向谁，解决什么问题？

Then follow with one question at a time:

• 第一版最核心的用户操作是什么？
• 用户为什么会用它，而不是继续用现有方案？
• 第一版必须做哪些功能？哪些功能明确后面再做？
• 项目里有哪些核心业务对象需要记录？
• 预期的数据量和用户量级大概是什么规模？（首年几百人试用，还是公开上线几万人？核心数据大概多少条？这会影响后面的数据库选型和索引设计）
• 这个项目更适合先做网页、小程序、App、后台系统，还是纯后端服务？

Discussion Guardrail

Explicitly tell the agent:

当前阶段只做产品讨论和立项设计，暂时不要写代码。你要以产品合伙人和架构顾问的角色追问、补全盲区、收敛 MVP，不要进入代码生成模式。

Generate Outputs

After confirmation, generate:

specs/project-brief-spec.md (English)

Include:

• Project overview
• Target users
• Core problem
• Product value and differentiation
• MVP scope
• Explicit non-goals for v1
• User journey
• Core business objects
• Initial product form recommendation
• Milestones
• Acceptance criteria
• Open questions
• AI agent constraints

ai-rules/ updates

Generate or update platform rules:

• AGENTS.md for Codex
• CLAUDE.md for Claude Code
• .cursorrules for Cursor

Rules must include:

• Do not write implementation code before project brief is confirmed
• Always check specs/project-brief-spec.md before planning features
• Do not add v2/later features into v1 without explicit approval
• Update the brief when product scope changes

ai-rules/prompt-templates.md

Add prompts for:

• product clarification
• MVP boundary review
• competitor/reference analysis
• user flow review
• acceptance criteria audit

Offer Next Stage

After writing the brief:

立项文档已完成。下一步建议使用 ai-tech-advisor 选择技术栈。是否继续？

Do not start technical implementation unless the user asks to continue.

Red Flags

Common Mistakes

1. Starting with code before product scope is clear
2. Letting AI add attractive but non-MVP features
3. Not documenting what v1 explicitly will not do
4. Skipping business objects until database design
5. Treating the brief as final instead of versioned and updatable