writing-prds
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseWriting PRDs
撰写PRD
Scope
范围
Covers
- Turning a product idea into a decision-ready PRD with unambiguous scope, requirements, and success metrics
- Optionally producing a PR/FAQ (press release + FAQ) to force customer-centric narrative first
- For AI features: adding a Prompt Set + Eval Spec so “requirements” are testable and continuously checkable
When to use
- “Write a PRD / product spec / requirements doc for this feature.”
- “Turn these messy notes into a PRD we can align on.”
- “Create a PR/FAQ and then a PRD.”
- “This is an AI feature; I need evals + prompts to define behavior.”
When NOT to use
- You’re still choosing what strategy/market to pursue (do product vision / strategy first)
- You need discovery from scratch (research plan, problem validation) more than requirements
- You need a detailed engineering design doc (APIs, schemas, low-level architecture)
- You’re prioritizing among many initiatives (do roadmap prioritization first)
涵盖内容
- 将产品想法转化为一份可用于决策的PRD,包含明确的范围、需求和成功指标
- 可选生成PR/FAQ(新闻稿+常见问题),优先构建以客户为中心的叙事
- 针对AI功能:添加Prompt Set + Eval Spec,让“需求”可测试且能持续验证
适用场景
- “为这个功能撰写PRD/产品规格/需求文档。”
- “把这些零散的笔记整理成一份我们可以对齐的PRD。”
- “先创建PR/FAQ,再写PRD。”
- “这是一个AI功能;我需要评估指标和提示词来定义其行为。”
不适用场景
- 你仍在选择要追求的战略/市场(先做产品愿景/战略)
- 相比需求,你更需要从零开始的探索(研究计划、问题验证)
- 你需要详细的工程设计文档(API、schema、底层架构)
- 你需要在多个项目中进行优先级排序(先做路线图优先级排序)
Inputs
输入信息
Minimum required
- Product + target user/customer segment
- Problem statement + why now (what changed, what’s broken, or what opportunity exists)
- Goal(s) + non-goal(s) + key constraints (timeline, policy/legal, platform, dependencies)
- Success metric(s) + 2–5 guardrails (quality, safety, cost, latency, trust)
If it’s an AI feature (additionally)
- What the model/system should do vs must never do (policy + safety)
- Concrete examples of desired and undesired outputs
- How correctness will be evaluated (offline tests, human review, online metrics)
Missing-info strategy
- Ask up to 5 questions from references/INTAKE.md.
- If answers are still missing, proceed with clearly labeled assumptions and provide 2–3 options (scope, metric, rollout).
最低必填项
- 产品+目标用户/客户群体
- 问题陈述+为什么是现在(发生了什么变化、哪里出了问题,或者存在什么机会)
- 目标+非目标+关键约束(时间线、政策/法律、平台、依赖项)
- 成功指标+2–5个护栏指标(质量、安全、成本、延迟、信任)
如果是AI功能(额外要求)
- 模型/系统应该做什么和绝对不能做什么(政策+安全)
- 期望和非期望输出的具体示例
- 如何评估正确性(离线测试、人工审核、线上指标)
缺失信息处理策略
- 从references/INTAKE.md中提出最多5个问题。
- 如果仍缺少答案,继续推进并明确标注假设,同时提供2–3个选项(范围、指标、上线计划)。
Outputs (deliverables)
输出成果(交付物)
Produce a PRD Pack in Markdown (in-chat; or as files if the user requests):
- Context snapshot (what decision we’re making, constraints, stakeholders)
- Artifact selection (PR/FAQ vs PRD vs AI add-ons)
- PR/FAQ (optional) — customer narrative + FAQs
- PRD — goals/non-goals, requirements (R1…Rn), UX flows, metrics, rollout
- AI Prompt Set (if AI) — versioned prompts + examples + guardrails
- AI Eval Spec (if AI) — acceptance tests + judge prompts + pass/fail criteria
- Risks / Open questions / Next steps (always included)
Templates: references/TEMPLATES.md
生成一份Markdown格式的PRD包(在对话中;或根据用户请求生成文件):
- 上下文快照(我们要做的决策、约束、相关利益方)
- 产出物选择(PR/FAQ vs PRD vs AI附加组件)
- PR/FAQ(可选)——客户叙事+常见问题
- PRD——目标/非目标、需求(R1…Rn)、用户流程、指标、上线计划
- AI Prompt Set(如果是AI功能)——带版本的提示词+示例+护栏指标
- AI Eval Spec(如果是AI功能)——验收测试+评审提示词+通过/失败标准
- 风险/未解决问题/下一步行动(始终包含)
模板:references/TEMPLATES.md
Workflow (8 steps)
工作流程(8个步骤)
1) Decide the artifact set (don’t over-document)
1) 确定产出物集合(不要过度文档化)
- Inputs: User request + constraints.
- Actions: Choose: PR/FAQ only, PRD only, PR/FAQ → PRD, or PRD + AI add-ons (Prompt Set + Eval Spec).
- Outputs: Artifact selection + rationale.
- Checks: The artifacts match the decision being made and the audience.
- 输入: 用户请求+约束条件。
- 行动: 选择:仅PR/FAQ、仅PRD、PR/FAQ→PRD,或PRD+AI附加组件(Prompt Set + Eval Spec)。
- 输出: 产出物选择+理由。
- 检查: 产出物与要做的决策和受众匹配。
2) Intake + clarify decision and success
2) 收集并明确决策与成功标准
- Inputs: references/INTAKE.md.
- Actions: Ask up to 5 questions; confirm decision owner, timeline, constraints, and success metrics/guardrails.
- Outputs: Context snapshot.
- Checks: You can state “what we’re deciding” and “how we’ll measure success” in 1–2 sentences.
- 输入: references/INTAKE.md。
- 行动: 提出最多5个问题;确认决策负责人、时间线、约束条件以及成功指标/护栏指标。
- 输出: 上下文快照。
- 检查: 你可以用1–2句话说明“我们要做什么决策”和“我们如何衡量成功”。
3) Write the customer narrative first (PR/FAQ or PRD narrative)
3) 先撰写客户叙事(PR/FAQ或PRD叙事部分)
- Inputs: Context snapshot.
- Actions: Draft a customer-centric narrative (problem → solution → why now). If using PR/FAQ, draft the press release headline/summary and top FAQs.
- Outputs: Narrative section (and PR/FAQ if selected).
- Checks: A stakeholder can restate the customer benefit and urgency without jargon.
- 输入: 上下文快照。
- 行动: 起草以客户为中心的叙事(问题→解决方案→为什么是现在)。如果使用PR/FAQ,起草新闻稿标题/摘要和核心常见问题。
- 输出: 叙事部分(以及选中的PR/FAQ)。
- 检查: 相关利益方可以不用行话复述客户价值和紧迫性。
4) Lock scope boundaries (goals, non-goals, out of scope)
4) 锁定范围边界(目标、非目标、超出范围内容)
- Inputs: Narrative + constraints.
- Actions: Define goals, non-goals, and explicit exclusions; call out dependencies and assumptions.
- Outputs: Scope section(s) in the PRD.
- Checks: “What we are NOT doing” is as clear as what we are doing.
- 输入: 叙事+约束条件。
- 行动: 定义目标、非目标和明确的排除项;标注依赖项和假设。
- 输出: PRD中的范围部分。
- 检查: “我们不做什么”和“我们做什么”一样清晰。
5) Convert scope into testable requirements (R1…Rn)
5) 将范围转化为可测试的需求(R1…Rn)
- Inputs: Goals + user journeys.
- Actions: Write numbered requirements with acceptance criteria, edge cases, and non-functional needs (privacy, latency, reliability). Mark “must/should/could”.
- Outputs: Requirements table/list.
- Checks: An engineer or QA can turn requirements into test cases without asking you to interpret intent.
- 输入: 目标+用户旅程。
- 行动: 撰写带验收标准、边缘案例和非功能性需求(隐私、延迟、可靠性)的编号需求。标注“必须/应该/可以”。
- 输出: 需求表格/列表。
- 检查: 工程师或QA无需询问意图,就能将需求转化为测试用例。
6) Define UX flows + instrumentation plan
6) 定义用户流程+埋点计划
- Inputs: Requirements + current product surfaces/events.
- Actions: Describe key user flows/states; specify success metrics, guardrails, and event/data needs (what to log, where, who owns).
- Outputs: UX/flows section + metrics & instrumentation section.
- Checks: Every goal has at least one measurable metric and a realistic data source.
- 输入: 需求+现有产品界面/事件。
- 行动: 描述关键用户流程/状态;明确成功指标、护栏指标以及事件/数据需求(要记录什么、在哪里记录、负责人是谁)。
- 输出: 用户流程部分+指标与埋点部分。
- 检查: 每个目标至少有一个可衡量的指标和真实的数据来源。
7) If AI feature: ship prompts + evals as “living requirements”
7) 如果是AI功能:将提示词+评估规范作为“活需求”交付
- Inputs: Requirements + examples.
- Actions: Create a versioned Prompt Set and an Eval Spec (judge prompts + test set + pass thresholds). Include red-team/failure modes.
- Outputs: Prompt Set + Eval Spec drafts.
- Checks: The eval suite can fail when behavior regresses and pass when requirements are met.
- 输入: 需求+示例。
- 行动: 创建带版本的Prompt Set和Eval Spec(评审提示词+测试集+通过阈值)。包含红队测试/失败场景。
- 输出: Prompt Set + Eval Spec草稿。
- 检查: 当行为退化时评估套件会失败,当需求满足时会通过。
8) Quality gate + finalize for circulation
8) 质量把关+定稿分发
- Inputs: Full draft pack.
- Actions: Run references/CHECKLISTS.md and score with references/RUBRIC.md. Add Risks/Open questions/Next steps.
- Outputs: Final PRD Pack (shareable as-is).
- Checks: Decisions, owners, metrics, and open questions are explicit.
- 输入: 完整的草稿包。
- 行动: 使用references/CHECKLISTS.md检查,并通过references/RUBRIC.md评分。添加风险/未解决问题/下一步行动。
- 输出: 最终PRD包(可直接分享)。
- 检查: 决策、负责人、指标和未解决问题都明确清晰。
Quality gate (required)
质量把关(必填)
- Use references/CHECKLISTS.md and references/RUBRIC.md.
- Always include: Risks, Open questions, Next steps.
- 使用references/CHECKLISTS.md和references/RUBRIC.md。
- 必须包含:风险、未解决问题、下一步行动。
Examples
示例
Example 1 (B2B SaaS feature): “Write a PR/FAQ + PRD for ‘Saved views’ in our analytics dashboard for admins.”
Expected: PR/FAQ narrative, a scoped PRD with R1…Rn, metrics/guardrails, and a rollout plan.
Expected: PR/FAQ narrative, a scoped PRD with R1…Rn, metrics/guardrails, and a rollout plan.
Example 2 (AI feature): “Write a PRD + Prompt Set + Eval Spec for an ‘AI email reply’ assistant with brand tone constraints.”
Expected: requirements that include safety/brand constraints, a prompt set with examples, and an eval spec with judge prompts + pass/fail thresholds.
Expected: requirements that include safety/brand constraints, a prompt set with examples, and an eval spec with judge prompts + pass/fail thresholds.
Boundary example: “Write a PRD for ‘make onboarding better’ (no product context).”
Response: ask the minimum intake questions; if context remains missing, produce 2–3 scoped options + assumptions and recommend discovery before committing to requirements.
Response: ask the minimum intake questions; if context remains missing, produce 2–3 scoped options + assumptions and recommend discovery before committing to requirements.
示例1(B2B SaaS功能): “为我们的分析仪表板中面向管理员的‘保存视图’功能撰写PR/FAQ + PRD。”
预期产出:PR/FAQ叙事、带R1…Rn的限定范围PRD、指标/护栏指标,以及上线计划。
预期产出:PR/FAQ叙事、带R1…Rn的限定范围PRD、指标/护栏指标,以及上线计划。
示例2(AI功能): “为带有品牌调性约束的‘AI邮件回复助手’撰写PRD + Prompt Set + Eval Spec。”
预期产出:包含安全/品牌约束的需求、带示例的提示词集,以及带评审提示词+通过/失败阈值的评估规范。
预期产出:包含安全/品牌约束的需求、带示例的提示词集,以及带评审提示词+通过/失败阈值的评估规范。
边界示例: “为‘优化新用户引导’撰写PRD(无产品上下文)。”
回应:提出最低要求的收集问题;如果仍缺少上下文,生成2–3个限定范围的选项+假设,并建议在确定需求前先进行探索研究。
回应:提出最低要求的收集问题;如果仍缺少上下文,生成2–3个限定范围的选项+假设,并建议在确定需求前先进行探索研究。