文档

飞书产品需求交付

试用

Convert a lightweight product request plus related signed-in pages into a product-confirmed requirements baseline published as one Feishu document. Use when a product manager asks Codex to investigate existing pages, clarify roles and business rules, capture complete route-aware screenshots, define

它能做什么

Convert a lightweight product request plus related signed-in pages into a product-confirmed requirements baseline published as one Feishu document. Use when a product manager asks Codex to investigate existing pages, clarify roles and business rules, capture complete route-aware screenshots, define user stories and exception scenarios, obtain product confirmation, and write the final requirement to Feishu for downstream development and testing. Also use for requests such as “按现有页面出需求”“把需求写到飞书”“补齐角色、用户故事、异常场景和验收标准”.

技能文档

Product Requirement Delivery

Turn a lightweight requirement card into one clear, reviewable Feishu requirement document. Keep the product manager responsible for business decisions. Do not require the product manager to prescribe code, APIs, database design, development plans, or test plans.

Required workflow

1. Audit the intake

Read references/intake-card.md. Extract the requirement name, related pages, current problem, target users, desired result, known rules, exclusions, and references.

Do not ask the user to write a full PRD. Discover page facts independently. Propose low-risk business defaults. Ask only questions whose answers materially change roles, permissions, visible behavior, state changes, data meaning, exception handling, or acceptance results.

2. Investigate the real product

Open every related page and inspect its navigation, fields, filters, states, operations, terminology, permissions visible to the current account, empty states, and adjacent flows. Use the user's signed-in Chrome session when login state matters, after reading the applicable Chrome/browser skill.

Read and obey references/ui-evidence-rules.md. For every page, capture both route evidence and complete page evidence. Preserve an unannotated original; annotations are supplemental.

Classify statements as:

  • Observed: directly verified on the live page.
  • User stated: supplied by the user.
  • Recommended: proposed for product confirmation.
  • Unverified: inaccessible or not provable from the current page state.

Never fabricate a replacement interface. If a state is unavailable, define the required business behavior and mark placement or current behavior unverified.

3. Draft the requirements baseline

Create -需求基线-v0.1.md using references/fact-source-schema.md. The local Markdown is the auditable source used to publish the Feishu document.

The draft must include:

  • roles and permission boundaries;
  • core user stories and preconditions;
  • explicit normal-flow rules;
  • state, default, editability, visibility, effective-time, and historical-data rules;
  • exception and boundary scenarios;
  • scope in/out;
  • observable Given/When/Then acceptance criteria;
  • complete URLs and screenshot evidence.

Use stable IDs: ROLE-*, US-*, REQ-*, EX-*, AC-*, and UV-*. Every AC must reference at least one REQ, US, or EX.

Run:

python scripts/validate_fact_source.py  --expected-status draft

4. Audit ambiguity before review

For every operation, make the following answerable without guessing:

  1. Which role can see or perform it?
  2. What preconditions and business states allow it?
  3. What is the default value and which values are legal?
  4. What exactly triggers success, and when does the result become visible?
  5. What changes, what remains unchanged, and how is history treated?
  6. What happens for no permission, invalid state, empty/not-found data, duplicate action, stale or concurrent modification, timeout, partial batch failure, and retry?
  7. What feedback does the user receive on success and failure?
  8. Which behavior is explicitly out of scope?

Do not use vague expressions such as “按需”“适当”“视情况”“正常处理” unless they are immediately converted into objective conditions and results.

5. Stop for product confirmation

Present only unresolved business decisions, each with a recommended default and impact. Do not publish the final Feishu document until the product manager explicitly confirms the rules.

6. Freeze the baseline

After confirmation:

  • change status to 已确认(可发布到飞书);
  • set version to v1.0 or later;
  • record confirmation date and decisions;
  • remove tentative language from confirmed rules;
  • keep inaccessible page facts as Unverified without weakening confirmed business behavior.

Run the validator with --expected-status confirmed.

7. Publish one Feishu requirement document

Read references/deliverable-contract.md. Then read and follow the lark-doc skill and its required shared authentication instructions.

  • If the user provides an existing Feishu document, update that document.
  • Otherwise create one document titled -产品需求-v in the user-specified folder, or the authenticated user's default document location when no folder is specified.
  • Publish the confirmed source as readable Feishu content, not as an attached DOCX.
  • Upload and insert the original complete screenshots and, when useful, annotated copies.
  • Put the copyable full URL next to every screenshot group even when the browser address bar is visible.
  • Do not create development-Agent packages, test-Agent packages, interface specifications, database designs, or implementation checklists unless the user separately asks for them.

8. Read back and verify

Fetch the published Feishu document and verify:

  • title and version;
  • all required headings and IDs;
  • roles, user stories, rules, exceptions, scope, and ACs;
  • every associated page URL;
  • every required screenshot is embedded and readable;
  • no confirmed rule reverted to tentative wording.

Create an audit manifest from references/audit-manifest.example.json, save the fetched Markdown, and run:

python scripts/audit_delivery.py --manifest  --source  --published 

Fix the Feishu document and repeat the read-back audit until it passes.

Non-negotiable gates

  • One confirmed requirements baseline governs the Feishu document.
  • No final Feishu publication before product confirmation.
  • Every involved role has an explicit permission boundary.
  • Every core business goal has at least one user story and one acceptance criterion.
  • Normal, state, permission, empty/not-found, duplicate/retry, concurrency/stale-data, failure, and batch-partial-failure behavior are either defined or explicitly marked not applicable with a reason.
  • Route evidence includes the browser address bar; page evidence covers the complete relevant page with continuous context.
  • Screenshots never replace copyable full URLs.
  • No fabricated product UI or unverified technical implementation presented as fact.
  • No mandatory development or test package is produced from this skill.
  • Preserve the user's existing files and unrelated workspace changes.

Handoff

Lead with the Feishu document link and confirmed version. State any remaining Unverified page facts and what evidence was unavailable. Do not expose internal work files unless requested.

相关技能

面向产品经理和产品协作者的产品需求评审沙盘工作流。适用于澄清模糊需求、分析业务问题、共创产品需求文档、准备需求评审材料、模拟研发/设计/数据/运营/合规/业务负责人/用户等多角色质询、收敛一期范围、梳理指标、流程、边界、风险、依赖和兜底方案。支持 Markdown、PDF、飞书文档三种交付形态,默认交付 PDF。...

12 次安装1 星标

Use for creating, changing, reviewing, reverse-engineering or accepting requirements, PRDs, prototypes, competitor material or existing systems, including any small UI, field, column, tab, dropdown or legacy-HTML change. Supports /ads, /dig, /prd and /proto intent shortcuts where the host routes the

4 次安装1 星标

帮助 PM 在需求启动阶段,通过三步对话法(苏格拉底→第一性→奥卡姆)拆清问题本质, 产出可进 review 的 PRD v3.x 终稿。支持按需裁剪章节、按需启用场景定位/Agent 设计/学习循环等可选阶段。 输出优先飞书文档,降级本地 MD。

1 次安装

机票产品需求评审 Agent。对前端、后端、运营类机票需求文档进行结构化评审打分,含独立的流程图治理评审(去If化、卫语句、决策表、FSM、泳道图、分层建模)。输入需求文档(markdown/图片/HTML),输出分项评分、评语和改进建议。使用场景:用户说"评审需求"、"审需求文档"、"给需求打分"、"需求评审"...

17 次安装

Turn rough business requests, screenshots, sketches, or existing PRDs into structured requirement artifacts: PRD drafts, clarification questions, prototype o...

8 次安装

产品需求文档生成:从产品定位到功能定义、技术方案、商业化设计、路线图,输出结构化PRD文档。Invoke when user asks 写PRD、产品需求文档、产品方案、产品定义、产品立项.

1 次安装