Write PRDs and product specs with AI
Skills used in this guide (3)
产品经理必备利器！
Download pack → https://github.com/anbeime/skill/tree/main/skills/product-manager-toolkit/product-manager-toolkit把 PRD 意图、路线图诉求或产品讨论翻译成可实现的能力方案，先暴露约束、不变量、接口与待定决策，再开始跨服务开发。
npx skills add affaan-m/ECC -g当你有多步任务的规格或需求、动代码前使用。
npx skills add obra/superpowers -g
The pain of PRD isn't "wrong format" — it's "goal and acceptance unclear, dev guesses". The point: give background, target user, problem; let AI output "feature list + acceptance + edges", not decide "why build" for you.
Who this is for / not for
For:
- PMs shipping requirement docs fast
- Break vague ideas into verifiable features
- Team wants consistent PRD structure
Not for:
- Don't know the problem yet (AI guesses wrong)
- Ask AI to set business goals (you set direction)
- Publish to eng without review (review is yours)
Before you start
- Background: why, for whom, what pain.
- Constraints: timeline, resources, tech limits.
- (Optional) existing doc: AI extends it.
Steps
1. Install the skill
product-manager-toolkit is a PM skill (runs locally). Copy its folder into your client's skills directory (same as any Agent Skill):
| Client | Directory |
|---|---|
| Claude Code | ~/.claude/skills/ |
| Codex | ~/.codex/skills/ |
| Cursor | ~/.cursor/skills/ |
| Trae | ~/.trae/skills/ |
| CodeBuddy | ~/.codebuddy/skills/ |
| Tongyi Lingma | ~/.lingma/skills/ |
2. Background, then structure
Write a PRD for "merchant bulk price edit". Background: ops manually edits thousands of SKUs weekly, slow & error-prone.
Standard structure: background/goal/user-story/features/acceptance/edges&exceptions/tracking,
each feature with verifiable acceptance (Given-When-Then).3. Break capability with product-capability
Complex features → product-capability into capability points, map to acceptance:
Break "bulk price edit" into 3 capabilities: upload-validate, preview-diff, batch-execute.
One normal + one exception acceptance each.4. You review & finalize
Add business goal and priority, delete fabricated features, keep real ones.
Pitfalls
| Pit | Symptom | Fix |
|---|---|---|
| Goal unclear | Dev guesses | Write background & problem first |
| Vague accept | "just works" | Given-When-Then |
| Miss exceptions | Break in prod | Exception flow per feature |
| Fabricated | Wrote undone | You delete |
| No priority | All wanted | P0/P1/P2 |
FAQ
No background?
Hard. Give "why, for whom" or AI drifts.
Acceptance format?
Given-When-Then; verifiable beats adjectives.
Too many features?
Break capabilities, rank P0/P1/P2.
AI sets business goal?
No — you set direction; AI structures & fills.
Need review?
Yes — add goal & priority, delete fabricated parts.
Related guides
Continue across these scenes:
- Turn an outline into a deck people can actually present
- Brainstorm and mind-map with AI
- Write technical blogs and long-form with AI
相关源码
- anbeime/skill · product-manager-toolkit source
- affaan-m/ECC · product-capability source
- obra/superpowers · writing-plans source
- Agent Skill open format spec
一句话结论
结论：PRD 先给背景目标再让 AI 出"功能+验收+边界"，商业方向你定、评审你做。
References
Related Scenes
Translate and localize docs with AI
Machine translation's scare is "fluent but distorted" — wrong terms, wrong tone, stiff local phrasing. The point: give a glossary and target-reader tone, have it "translate + keep proper nouns", then you verify key sentences — not trust the whole blindly.
Write weekly reports and standups with AI
The pain of weekly reports isn't "can't write" — it's "turning a log into outcomes". The point: give AI the raw material (what you did, what's stuck), have it close into "progress / risk / next", not invent content for you.
Turn meeting recordings into a task list with AI
The biggest waste in meetings isn't the meeting — it's nobody remembering "who, by when, what" afterward. The point of this flow is to compress scattered discussion into a task table with owners and due dates, not to write another minutes doc nobody reads.
Handle your inbox and support messages with AI
The red line for email automation is "draft yes, auto-send careful". "Triage + draft + you hit send" is safest; letting AI blast-send means you own the mistake, not it.