AI PRD generation, PRD review, and Feishu/Lark collaboration workflows for B2B product management and system design.
中文简介：这是一套面向 B 端产品经理的 AI Native 工具链，把 写 PRD、审 PRD、在飞书里协作收口 串成一个真正可落地的闭环。
PRD Writing Toolkit is designed for teams that want more than a single prompt file. It turns product writing and review into a repeatable operating system:
- AI PRD generation for turning raw context into structured drafts
- PRD review for catching gaps, contradictions, and risk before handoff
- Feishu/Lark collaboration for writing comments back, collecting decisions, and controlling a real delivery gate
- B2B product management and system design workflows that respect complexity instead of forcing every requirement into the same template
This toolkit is an integrated workflow product, but it does not erase where the core upstream skill work came from.
create-prd-skillin this repository builds on the original pmYangKun/create-prd-skillcheck-prd-skillin this repository builds on the original pmYangKun/check-prd-skill
Huge thanks to pmYangKun for creating and open-sourcing the original create-prd and check-prd skill repositories. PRD Writing Toolkit extends those foundations into a fuller end-to-end workflow with a vendored quality framework and a Feishu/Lark delivery gate.
- Product managers writing B2B PRDs, solution specs, and internal system proposals
- AI-native PM teams that want a reusable workflow, not just one-off prompting
- Product ops or design review teams that need a consistent delivery gate
- Teams using Feishu/Lark as the place where review feedback actually gets resolved
Start from a business brief, workshop notes, or even a one-line idea. The generator classifies the request by complexity, chooses the right chapter depth, and produces a structured PRD scaffold instead of a generic wall of text.
Best for:
- New feature PRDs
- Internal system specs
- Module-level solution drafts
- Turning meetings into first-pass product docs
Review a PRD using the right path for the document you actually have:
- Path A for standard, chapter-based PRDs
- Path B for messy, free-form docs that still need rigorous review
This keeps review quality high even when documentation quality is uneven.
Best for:
- Pre-dev readiness review
- Cross-functional PRD critique
- System design quality checks
- Catching missing flows, data, permissions, and exception handling
Push findings back into Feishu/Lark, collect 采纳 / 不采纳, generate confirmation cards before any document edit, log waivers, and decide whether the PRD is truly ready for development.
Best for:
- PRD review in Feishu docs
- PM/design/dev alignment workflows
- Auditability for review decisions
- Teams that want a real delivery gate, not just comments scattered across threads
The production prompt, structured diagram data, and exported assets for this workflow live in assets/diagrams/.
- Complexity-aware by default: L1-L4 classification prevents over-writing tiny changes and under-specifying large systems.
- Chapter-first, not checklist-only: when a PRD has structure, review follows that structure. When it does not, the fallback 14-dimension path keeps quality from collapsing.
- Delivery-gate oriented: the Feishu assistant does not stop at “leave comments”; it tracks replies, confirmations, waivers, and readiness.
- Progressive loading: skills load only the framework files needed for the current stage, which keeps large workflows more stable in practice.
- Built for B2B product work: this toolkit assumes business flows, permissions, exceptions, architecture, and rollout matter.
git clone https://github.com/Scofy0123/prd-writing-toolkit.git
cd prd-writing-toolkitThe bundled skills in this repo already include vendored framework snapshots for zero-dependency installation. If you also want the framework available as a standalone reference module in your local skill inventory, install the top-level prd-quality-framework too:
cp -R prd-quality-framework ~/.claude/skills/prd-quality-frameworkbash create-prd-skill/install.sh
bash check-prd-skill/install.shThis workflow requires lark-cli plus user-authenticated access to the target Feishu/Lark documents.
cp -R feishu-prd-review-loop ~/.agents/skills/feishu-prd-review-loopExamples:
/create-prdwith a business brief to generate the first draft/check-prdwith a PRD to run a readiness review- Share a
feishu.cndoc link and ask for a review loop when you want anchored comments and a delivery decision
The Feishu assistant is intentionally more opinionated than a normal commenting bot:
- It classifies the PRD scope with L1-L4 complexity
- It chooses the actual review chapter set
- It writes one issue per anchored comment
- It reads reply threads and normalizes
采纳 / 不采纳 - It requires PM confirmation before editing the PRD
- It logs replies, waivers, and final processing into a feedback base
- It evaluates a delivery gate instead of assuming “comments resolved = ready”
That makes it useful for real team workflows where review history matters.
| Module | Purpose | Upstream Origin |
|---|---|---|
create-prd-skill/ |
Complexity-aware AI PRD generation | Based on pmYangKun/create-prd-skill |
check-prd-skill/ |
Chapter-based review with fallback dimensions | Based on pmYangKun/check-prd-skill |
feishu-prd-review-loop/ |
Feishu/Lark review-to-delivery workflow | Workflow variant aligned with the check-prd-skill Feishu collaboration branch |
prd-quality-framework/ |
Shared L1-L4 framework, chapter standards, and global checks | Packaged here as the first-class foundation of the toolkit |
- Improve standalone installers for one-command setup
- Add more eval cases for L1/L2 lightweight requirements
- Expand English-language examples for global discoverability
- Add more operational templates for review governance and handoff quality
- Improve chapter standards inside
prd-quality-framework/ - Add richer example prompts and PRD samples
- Contribute Feishu/Lark workflow fixtures and safer integration tests
- Open issues for gaps in AI PRD generation, PRD review, or delivery gate behavior