AgentStack
SKILL verified MIT Self-run

Vibe Coding Prd

skill-yangjinmin666-vibe-coding-prd-skill-vibe-coding-prd · by yangjinmin666

Turn fuzzy ideas, clear requirements, raw engineering PRDs, Feishu/Lark docs, competitor analysis, research notes, screenshots, or mixed product material into a concise but complete Markdown PRD that can be sent directly to Claude Code, Codex, or another coding agent. Use when the user says "帮我写一个用来vibe coding的PRD", asks for a Vibe Coding PRD, wants to transform requirements into coding-agent-rea…

No reviews yet
0 installs
16 views
0.0% view→install

Install

$ agentstack add skill-yangjinmin666-vibe-coding-prd-skill-vibe-coding-prd

✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.

Are you the author of Vibe Coding Prd? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Vibe Coding PRD

Create an implementation-ready PRD for coding agents. Keep it concise, but do not omit decisions that affect scope, architecture, data, AI workflow, UI states, failure handling, or acceptance.

Non-Negotiable Rules

  • Do not guess core intent. If user, scenario, pain point, product shape, MVP scope, technical direction, or wireframe is unclear, ask before continuing.
  • Use modal/user-input confirmation tools when available. If unavailable, ask numbered choice questions in chat.
  • Every blocking question must be option-based: 2-4 choices, one recommended choice when possible, and a short tradeoff note. Use open-ended questions only when choices would be misleading.
  • Move through the four confirmation gates in order: demand definition, feature/MVP priority, technical stack, core wireframe. Do not output the final PRD until all gates are confirmed or explicitly marked as assumed by the user.
  • Strip business persuasion, market sizing, vision slogans, organization background, and repeated narrative unless it changes implementation.
  • Preserve implementation signals: users, roles, workflows, pages, states, data objects, permissions, APIs/events, AI prompts/tools, evaluation, fallbacks, observability, and acceptance criteria.
  • For AI products, expose the Product Engineer route: client action, API/event entry, domain service, orchestrator, model/tool/retrieval dependency, data read/write, state transition, human review/fallback, trace/metric/audit, cost or release control.
  • Acceptance criteria must be observable and reproducible. Avoid vague criteria such as "体验良好", "更智能", "页面美观", or "性能优秀".
  • Default to the user's language. If the user writes in Chinese, ask and output in Chinese while preserving useful technical terms.
  • The skill produces PRDs, not product code. If the user asks to implement after the PRD, treat that as a separate coding task.

Workflow Map

  1. Classify and extract input: read [references/01-input-routing.md](references/01-input-routing.md).
  2. Clarify with gates: read [references/02-confirmation-gates.md](references/02-confirmation-gates.md).
  3. Design features and MVP: read [references/03-feature-mvp.md](references/03-feature-mvp.md).
  4. Recommend technical stack: read [references/04-tech-stack.md](references/04-tech-stack.md).
  5. Draft core wireframe: read [references/05-wireframe.md](references/05-wireframe.md).
  6. Write detailed modules: read [references/06-detailed-modules.md](references/06-detailed-modules.md).
  7. For AI products or AI-heavy modules: also read [references/07-ai-pe-agent-design.md](references/07-ai-pe-agent-design.md).
  8. Write acceptance criteria: read [references/08-acceptance-criteria.md](references/08-acceptance-criteria.md).
  9. Generate final PRD: read [references/09-final-prd-template.md](references/09-final-prd-template.md).

Read only the references needed for the current phase. If the user provides a raw PRD link, read the source material before summarizing. For Feishu/Lark Docx/Wiki links, use the available lark-doc or equivalent document connector/skill; if access fails, ask the user to grant access, export Markdown, or paste the content.

Input Classification

Classify the material as exactly one primary type:

  • 明确需求: already includes user, scenario, pain point, and product form/output.
  • 原始 PRD: long or structured product/engineering document, Feishu/Lark doc, Markdown, PDF, or PRD-like spec.
  • 竞品分析: mainly describes other products, screenshots, comparisons, flows, or feature inspiration.
  • 混合材料: multiple source types with possible conflicts.
  • 信息不足: missing user, scenario, pain point, or product shape.

State the classification briefly, what will be extracted, and whether confirmation is needed.

Confirmation Gates

Gate 1: Demand definition.

  • 给谁用
  • 什么场景
  • 解决什么问题
  • 什么产品形态
  • 核心输入/输出
  • 明确不做

Gate 2: Feature list and priority.

  • 一级模块、二级功能、功能描述
  • P0/P1/P2/Later/不做
  • MVP development order
  • dependencies and reason for priority

Gate 3: Technical stack.

  • one recommended plan and, only when useful, one simpler or more scalable alternative
  • explain tradeoffs in plain language for non-technical users
  • map important features to implementation choices

Gate 4: Core wireframe.

  • 1-3 core pages or interfaces
  • ASCII/text wireframes
  • key actions, states, empty/error/loading/success states
  • main click path

Final PRD Standard

The final Markdown PRD must let a coding agent start without guessing:

  • what to build and what not to build
  • first version scope and priority order
  • pages, modules, states, and flows
  • technical stack and architecture route
  • data entities, APIs/events, permissions, and integrations when relevant
  • AI agent workflow, prompts, tools, memory, evaluation, and fallback when relevant
  • acceptance criteria that can become tests or manual checks
  • suggested coding-agent execution order

If confirmation is incomplete, output a 待确认项 section and ask the next option-based question instead of pretending the PRD is final.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet — be the first.

Versions

  • v0.1.0 Imported from the upstream source.