RPD — Rapid Product Document

A structured product management skill covering the full project lifecycle.
覆盖项目全生命周期的结构化产品管理 Skill。

---

Language Detection / 语言检测

Detect the user's language from their first message and respond in the same language throughout the entire session.

• If the user writes in Chinese (中文) → respond in Chinese
• If the user writes in English → respond in English
• If unclear → default to Chinese, but ask: "你希望用中文还是英文？ / Would you prefer Chinese or English?"

All prompts, questions, and outputs should use the detected language. The .project-state.md file content follows the user's language.

---

Hard Constraints / 硬性红线

The following rules are absolutely不可违反 throughout the entire Skill execution:
以下规则在整个 Skill 执行过程中绝对不可违反：

1. Security scan不可跳过 / 安全扫描不可跳过：每次 Flow B（接手）和 Flow C（继续）开始时，必须先运行 security-scanner.py。If exit code 2, must hard block / 如果 exit code 2，必须硬阻断，不可降级。
2. State validation不可跳过 / 状态校验不可跳过：每次更新 .project-state.md 后，必须运行 state-validator.py。校验失败必须回滚。
3. Max 3 questions per round / 每轮问题不超过 3 个：诊断过程中，每轮最多问 3 个问题，等用户回答后再继续。
4. Don't skip concept PRD / 不跳过概念版 PRD：标准模式下，必须先输出概念版 PRD 并获得用户确认，再出落地版。
5. Don't overwrite without backup / 不覆盖未备份的文件：写入 .project-state.md 前，必须先运行 state-guard.py --action backup。
6. No path traversal / 路径不越界：所有文件操作必须在当前项目目录内，禁止 ../ 路径遍历。

---

When to Use / 何时使用

Positive matches / 正向匹配：

English	中文
User describes a new product idea or feature request	用户描述新产品想法或功能需求
"I want to build..." / "I have an idea..."	"我想做一个..."、"有个点子..."
"Take over this project" / "Analyze this project"	"接手这个项目"、"分析一下这个项目"
"Continue development" / "What's next?"	"继续开发"、"下一步做什么"、"接着做"
"Update progress" / "Sync status"	"更新进度"、"同步状态"

Negative matches / 反向排除（不使用本 Skill）：

• User is just writing code or debugging, no product direction questions
• User is doing pure technical discussion (architecture, performance, etc.)
• Project already has complete requirements docs and user has no direction confusion
• 用户只是在写代码或调试，没有产品方向的问题
• 用户在做纯技术讨论（架构选型、性能优化等）
• 项目已有完善的需求文档且用户没有方向困惑

---

Inputs Required / 所需输入

• User's natural language description (product idea, takeover intent, or continue development intent)
• Current working directory (for scanning project files)
• 用户的自然语言描述（产品想法、接手意图、或继续开发意图）
• 当前工作目录（用于扫描项目文件）

---

Procedure / 流程

Step 0: Intent Router / 意图路由

Deterministic routing (must execute) / 确定性路由（必须执行）：

Before any other action, run / 在做任何其他事情之前，先运行：

python scripts/intent-router.py "用户输入的原文"

Route based on recommended_flow field / 根据输出的 recommended_flow 字段决定走哪条路：

recommended_flow	Action / 动作
scene_exploration	Enter Scene Exploration Mode (see below) / 进入场景探索模式
warm_start	Warm start: ask 1 core question then re-assess / 暖启动：问 1 个核心问题后重新判断
standard_diagnosis	Standard 3-round diagnosis / 标准三轮诊断
flow_b	Take over half-finished / 接手半成品
flow_c	Continue development / 继续开发
ask_user	Uncertain, ask user / 不确定，主动问用户

Warm Start Layer / 暖启动层

Trigger Word Expansion / 触发词扩展表

User says (EN)	User says (ZH)	RPD understands as
"Help me make something"	"帮我做个东西"	New project (warm start)
"I have an idea"	"我有个想法"	New project (warm start)
"Um... just..."	"那个...就是..."	New project (warm start, proactively guide)
"Help me sort through this"	"帮我理一下"	Take over half-finished
"Halfway done"	"做到一半了"	Take over half-finished
"Continue from here"	"接着来"	Continue development

Warm Start Flow / 暖启动流程

1. Confirm with a friendly question / 用一句通俗话确认:

   • EN: "You want to start something from scratch, right?"
   • ZH: "你是想从零开始做一个新东西，对吧？"

2. First round: only ask 1 core question / 第一轮只问 1 个最核心的问题:

   • EN: "Who is this for? Yourself or others?"
   • ZH: "你这个东西是给谁用的？自己用还是给别人用？"

3. Assess user maturity based on answer / 根据回答判断用户成熟度:

   • Vague answer (e.g., "just... tracking expenses") / 回答模糊（"就是...记个账"）→ 降级为"场景探索模式", use everyday language to guide / 用生活化语言引导
   • Clear answer (e.g., "A budgeting tool for indie developers") / 回答清晰（"给独立开发者用的记账工具"）→ 进入标准三轮诊断

Core Principle / 核心原则: Talk about scenarios first, then people, then tech. Don't start with jargon like "target user" or "business model". / 先聊场景，再聊人，最后聊技术。不要一上来就甩"目标用户"、"商业模式"这些术语。

Scene Exploration Mode / 场景探索模式

Trigger / 触发条件: intent-router.py returns maturity: vague / 返回 maturity: vague

Goal / 目标: Use 3 rounds of casual conversation to help beginners turn vague ideas into concrete scenarios.
用 3 轮生活化对话，帮小白把模糊想法变成具体场景。

Round 1: Talk about scenarios / 第 1 轮：聊场景（问 2 个）

• EN: "When do you usually think about [doing this]?" / "What do you currently use to [solve this problem]?"
• ZH: "你平时什么时候会想 [做这件事]？" / "现在用什么方式 [解决这个问题]？"

Round 2: Talk about pain points / 第 2 轮：聊痛点（问 2 个）

• EN: "What annoys you most about [current solution]?" / "Is there something you want but don't have?"
• ZH: "现在 [这个方式] 最烦的是什么？" / "有没有什么想要但现在没有的？"

Round 3: Talk about basics / 第 3 轮：聊基础（问 2 个）

• EN: "Where is the data stored? (on your phone / server / cloud)" / "Do users need to register and login?"
• ZH: "数据存在哪里？（自己手机 / 服务器 / 云端）" / "需要用户注册登录吗？（是 / 否）"

After completion / 完成后:
AI summarizes the scenario in its own words and asks: "I understand, you want to make a [summary], right?"
AI 用自己的话总结场景，问"我理解了，你是想做一个 [总结]，对吧？"

After user confirms / 用户确认后:
Output concept PRD in plain language (no template jargon), skip standard 3-round diagnosis.
直接输出概念版 PRD（人话版，不用模板术语），跳过标准三轮诊断。

Intent Routing / 意图分发

Route based on user intent / 根据用户意图分发：

Intent (EN)	Intent (ZH)	Route to
"new idea", "I want to build...", "I have an idea"	"新想法"、"我想做一个"、"有个点子"	Flow A: New Project
"take over", "analyze project", "half-finished"	"接手"、"分析项目"、"这个项目做到一半"	Flow B: Takeover
"continue development", "what's next", "keep going"	"继续开发"、"下一步"、"接着做"	Flow C: Continue
Unclear / 不确定	→ Ask: "What would you like to do? A) Start a new project B) Take over an existing project C) Continue previous development"

Shortcut Commands / 快捷命令

Command	Action
/rpd new	New project flow
/rpd take	Take over half-finished project flow
/rpd cont	Continue development
/rpd status	Show current project status summary
/rpd scan	Run security scan + project scan only, don't enter flow

---

Flow A: New Project / 新项目

From idea to PRD with structured diagnosis.
从想法到 PRD 的结构化转化。

A1: Three-Perspective Diagnosis / 三视角诊断

Restate your understanding of the core need, then tell the user:

"I'll ask a few rounds of questions to align requirements. This will take about 3 minutes."
"接下来我会通过几轮提问对齐需求，总共大约需要 3 分钟。"

Round 1 (User Perspective / 用户视角) — pick 2-3:

English	中文
Who is the target user? (age, occupation, scenario)	目标用户是谁？（年龄段、职业背景、使用场景）
How do they currently solve this problem?	他们现在用什么方式解决这个问题？
What ONE specific thing does your product help with?	你的产品帮他解决的是哪一件具体的事？
How often will they use it? (daily / weekly / occasionally)	这个场景下用户会多久用一次？

After questions: "Round 1, 2 rounds remaining" / 「第 1 轮，还剩 2 轮」

Round 2 (Business Perspective / 商业视角) — ask all:

English	中文
Is this free or paid?	这是免费产品，还是用户要付费使用？
If paid, what model: one-time / subscription / freemium?	如果收费，模式是哪种：一次性买断 / 订阅 / 免费增值？
How long do you plan to work on this? (validation / long-term)	预期把这个产品做多久？（验证型项目 / 长期运营）

After questions: "Round 2, 1 round remaining" / 「第 2 轮，还剩 1 轮」

Round 3a (Technical - Infrastructure / 技术 - 基础设施) — ask all:

English	中文
Where is data stored? (on user's phone / server / cloud)	数据存在哪里？（自己手机 / 服务器 / 云端）
Do you need user registration and login? (yes / no)	需要用户注册登录吗？（是 / 否）
What devices will users primarily use? (mobile / desktop / both)	用户主要在什么设备上用？（手机 / 电脑 / 都要）

After questions: "Round 3a, 2 rounds remaining" / 「第 3a 轮，还剩 2 轮」

Round 3b (Technical - Scope / 技术 - 范围) — ask all:

English	中文
How many main pages will the product have?	产品预计有几个主要页面？
How long do you expect this version to take? (weeks)	这个版本预期几周内可以做出来？
Does it need to connect to other services? (payment, maps, etc.)	是否需要连接其他服务？（支付、地图等）

After questions: "Round 3b, 1 round remaining" / 「第 3b 轮，还剩 1 轮」

Round 4a (Security - Account & Data / 安全 - 账户与数据) — ask all:

English	中文
Does it involve user registration and login? → Account security needed	是否涉及用户注册和登录？→ 需要考虑账户安全
Does it have user-generated content (comments, posts, avatars)? → Content moderation needed	是否有用户发布内容的功能（评论、帖子、头像等）？→ 需要内容审核
Does it have file upload? → File security measures needed	是否有文件上传功能？→ 需要文件安全措施

After questions: "Round 4a (Security), 1 round remaining" / 「第 4a 轮（安全），还剩 1 轮」

Round 4b (Security - AI & Scale / 安全 - AI 与规模) — ask all:

English	中文
Does it use AI/LLM features? → Prompt protection needed	是否使用 AI/大模型功能？→ 需要 Prompt 保护
Does it involve payments or sensitive data? → Encryption and compliance needed	是否涉及支付或敏感数据？→ 需要数据加密和合规
Expected user scale? → Affects security investment priority	预期用户规模？→ 影响安全投入的优先级

After questions: "Round 4b (Security), last question!" / 「第 4b 轮（安全），最后一个问题了！」

Beginner Mode / 小白模式（当 intent-router 返回 maturity: vague 或 scene_exploration 时）：

安全视角由 AI 根据前几轮回答自动判断，不询问用户。
Security perspective is auto-determined by AI based on previous answers, without asking the user.

If previous rounds mentioned... / 如果前几轮提到...	Auto-add / 自动加入
"login"/"register"/"account" / "登录"/"注册"/"账号"	Account security / 账户安全
"comment"/"post"/"avatar"/"UGC" / "评论"/"发帖"/"头像"	Content moderation / 内容审核
"upload"/"attachment"/"image" / "上传"/"附件"/"图片"	File security / 文件安全
"AI"/"smart"/"bot" / "AI"/"智能"/"机器人"	Prompt protection / Prompt 保护
"payment"/"money"/"bank" / "支付"/"钱"/"银行卡"	Encryption & compliance / 加密合规
None of the above / 以上都没有	Skip security perspective / 跳过安全视角

In concept PRD, add one line at the end / 在概念版 PRD 末尾一句话带过：

Security considerations: [auto-filled items] / 安全考虑：[自动填入的项]

Rules / 规则：

• Max 3 questions per round, wait for answers before continuing
• Show progress after each round
• Use simple language for beginners, avoid jargon
• Do NOT output a full PRD at this stage
• 每轮最多 3 个问题，等待回答后再继续
• 每轮结束后显示进度
• 对小白用户使用通俗语言
• 此阶段禁止输出完整 PRD

A2: Concept PRD / 概念版 PRD

After diagnosis, output concept PRD (≤ 200 words):

English:

Who it's for: [one sentence]
Problem to solve: [one specific pain point]
How it works: [Web / App / Mini-program / CLI]
Tech requirements: [auth / data / third-party deps]
MVP Features (≤ 3):
  1. [Feature 1]
  2. [Feature 2]
  3. [Feature 3]
Out of Scope: [explicitly excluded features]
Business Model: [pricing or free]

中文：

给谁用的：[一句话描述目标用户]
解决什么问题：[一个具体痛点]
怎么用的：[Web / App / 小程序 / 命令行]
技术上需要什么：[账号体系 / 数据方案 / 第三方依赖]
最小可用版功能（≤ 3 条）：
  1. [功能一]
  2. [功能二]
  3. [功能三]
本版本不做：[明确排除的功能]
商业模式：[收费方式 或 免费]

Beginner Concept PRD / 小白版概念 PRD

When intent-router.py returns maturity: vague or scene_exploration, use this template:
当 intent-router.py 返回 maturity: vague 或 scene_exploration 时，使用此模板：

📱 What you want to make / 你想做的东西：
[One sentence, plain language / 一句话描述，用人话]

✨ Main features (do these 3 first) / 主要功能（先做这 3 个）：
1. [Feature 1, plain language / 功能一，用人话]
2. [Feature 2, plain language / 功能二，用人话]
3. [Feature 3, plain language / 功能三，用人话]

🚫 Not doing this time / 这次不做：
[Exclusions, plain language / 排除项，用人话]

💰 Free or paid / 免费还是收费：
[One sentence / 一句话]

🔧 Technical approach / 技术方案：
[One sentence, no jargon. E.g. "Data stored on your phone, no account needed"]
[一句话，不用术语。例如"数据存在你自己手机里，不需要注册账号"]

Then ask: "Is this direction aligned? Any adjustments needed?" / "方向是否对齐？有需要调整的地方吗？"

A3: Scope Freeze / 范围冻结

Output freeze checklist:

English:

Before we proceed to the full PRD, please confirm these won't change:

✅ Target User: [...]
✅ Core Features (≤ 3): [...]
✅ Platform: [...]
✅ Out of Scope: [...]

Please reply "confirmed" to continue.

中文：

在开始落地版之前，请确认以下内容不会再修改：

✅ 目标用户：[...]
✅ 核心功能（≤ 3 条）：[...]
✅ 平台选择：[...]
✅ 本版本不做：[...]

请回复"确认"后继续。

A4: Full PRD / 落地版 PRD

Refer to references/prd-template.md for the full template. Output complete PRD with all sections.

After completion, run the Six Blind Spots checklist (see references/prd-template.md).

PRD Completeness Check / PRD 完整性校验

After generating the full PRD, must execute / 生成落地版 PRD 后，必须执行：

python scripts/prd-validator.py <prd-file> --format text

If output shows FAIL, supplement missing content based on gaps list, then re-validate. / 如果输出 FAIL，根据 gaps 列表补充缺失内容后重新校验。

A5: Generate State File / 生成状态文件

Generate .project-state.md following references/state-file-spec.md.

Decision Table Format / 决策记录格式:

The decision table in .project-state.md must use 5 columns:
| 日期 | 类型 | 决策 | 原因 | 影响范围 |

Type values / 类型字段: 技术选型 / 架构模式 / 安全策略 / 业务逻辑 / 其他

State File Protection / 状态文件保护

After generating or updating .project-state.md, must execute / 生成或更新 .project-state.md 后，必须执行：

1. Backup (if file already exists) / 备份（如果文件已存在）：

   python scripts/state-guard.py .project-state.md --action backup

2. Validate / 校验：

   python scripts/state-validator.py .project-state.md

Output: "Project initialized, state file generated. Say 'continue development' to resume context." / "项目已初始化，状态文件已生成。对我说「继续开发」来恢复上下文。"

---

Flow B: Takeover Half-Finished Project / 接手半成品

Analyze existing code, generate PRD and state file.
分析已有代码，生成 PRD 和状态文件。

B1: Security Scan / 安全扫描

python scripts/security-scanner.py <project-directory>

If exit code 2 → output security report and STOP. / 输出安全报告并停止流程。

B2: Project Scan / 项目扫描

python scripts/project-scanner.py <project-directory>

Analyze JSON output for: tech stack, components, API routes, TODOs, stats.

B3: Confirm Inference / 推断确认

Based on scan results, confirm with user:

English:

1. "I infer this is a [X] product, correct?"
2. "Core features I understand: [list]. Anything missing?"
3. "Out of scope: [list]. Confirmed?"
4. "Tech stack is [X], correct?"

中文：

1. "我推断这是一个 [X] 产品，对吗？"
2. "核心功能我理解为这 3 个：[列出]，有遗漏吗？"
3. "本版本不做：[排除项]，确认吗？"
4. "技术栈是 [X]，对吗？"

B4: Generate PRD + State File / 生成 PRD + 状态文件

Generate simplified PRD (concept level is sufficient) and .project-state.md.

Feature progress auto-marking based on code analysis:

• Has complete implementation → ✅ Completed / ✅ 已完成
• Has partial code → 🔨 In Progress / 🔨 进行中
• No code but in PRD → ⏳ Not Started / ⏳ 未开始

Run validation:

python scripts/state-validator.py .project-state.md

Output: "Project takeover complete, state file generated. Say 'continue development' to start." / "已接手项目，状态文件已生成。对我说「继续开发」开始下一步。"

---

Flow C: Continue Development / 继续开发

Cross-agent / cross-time context recovery.
跨 agent/跨时间的上下文恢复。

C1: Check State File / 检查状态文件

Check if .project-state.md exists:

• Not found → "No state file found. Please 'analyze project' or 'new project' first." / "没有找到项目状态文件，请先「分析项目」或「新建项目」"
• Found → continue

Review Mode / 回顾模式

When user has been inactive for >3 days, or explicitly says "回顾"/"之前做了什么"/"what did we do"/"review":
当用户超过 3 天未操作，或明确说"回顾"/"之前做了什么"时：

1. Output PRD summary in plain language / 输出 PRD 摘要（用人话）
2. Output feature list with current status / 输出功能清单和当前状态
3. Ask / 问："要继续开发，还是想改改之前的方案？"/ "Continue development, or revise the plan?"
4. If revise → re-enter diagnosis (Flow A1) / 如果改方案 → 重新进入诊断流程（Flow A1）
5. If continue → proceed to C2 / 如果继续 → 进入 C2

C2: Security Scan / 安全扫描

python scripts/security-scanner.py <project-directory> --state-file .project-state.md

If exit code 2 → output security report and STOP.

C3: State Validation / 状态校验

python scripts/state-validator.py .project-state.md

If validation fails → try to fix or prompt user to regenerate.

C4: Gap Analysis / 差距分析

python scripts/gap-analyzer.py <project-directory> --state-file .project-state.md

C5: Generate Action Plan / 生成行动建议

Based on gap analysis:

English	中文
Progress: X/Y features completed (Z%)	当前进度：X/Y 功能已完成（Z%）
Next: "Suggest completing [feature] because [reason]"	下一步建议："建议先完成 [功能]，因为 [原因]"
Blocker: "Blocker [X] unresolved, suggest [solution]"	阻塞提醒："阻塞项 [X] 尚未解决，建议 [方案]"
Deviation: "Found [X] deviates from PRD"	偏离警告："发现 [X] 模块偏离 PRD"
Estimate: "Completing next feature takes ~[time]"	预估工作量："完成下一个功能预计需要 [时间]"

C6: Post-Development Update / 开发完成后的状态更新

After development completes, remind user:

"Feature complete. Update project state file? (Updates enable more accurate progress tracking)"
"功能已完成。要更新项目状态文件吗？（更新后可以更准确地追踪进度）"

Before updating:

1. Check for concurrent modifications: git diff .project-state.md
2. If conflicts exist, merge first before updating
3. Update .project-state.md with new status and last-modified-by
4. Run validation

---

Token Budget Control / Token 预算控制

When token budget is tight, degrade gracefully:

Component	EN	ZH
Concept PRD	Always output (low cost, ≤ 200 words)	始终输出（成本低）
Full PRD	Output only feature list + page structure	只输出功能列表 + 页面结构
Continue Dev	Output only progress summary + next step	只输出进度摘要 + 下一步建议
Security Scan	Always execute (deterministic script, no LLM tokens)	始终执行（确定性脚本，不消耗 token）

Turbo Mode / 极速模式

Trigger conditions / 触发条件:

• User explicitly says "hurry up", "simpler", "don't ask so much" / 用户明确说"快一点"、"简单点"、"别问那么多"
• OR token remaining < 2000 / 或 Token 剩余 < 2000

Turbo mode behavior / 极速模式行为:

1. Skip concept PRD, output minimal landing version directly (only feature list + page structure) / 跳过概念版 PRD，直接出精简落地版（只含功能列表 + 页面结构）
2. Skip scope freeze confirmation (assume user has verbally confirmed) / 跳过范围冻结确认（假设用户已口头确认）
3. Continue development only outputs: "Progress X%, next: [specific task], estimate [time]" / 继续开发只输出："进度 X%，下一步：[具体任务]，预估 [时间]"
4. Security scan always executes (no token consumption) / 安全扫描始终执行（不消耗 Token）

Turbo mode output format / 极速模式输出格式:

📋 [Project Name] Status / [项目名] 状态
- Progress: X/Y (Z%) / 进度：X/Y（Z%）
- Next: [one sentence] / 下一步：[一句话]
- Blocker: [one sentence or None] / 阻塞：[一句话 或 无]
- Estimate: [time] / 预估：[时间]

---

Verification / 验证

After every state file change:

python scripts/state-validator.py .project-state.md

Before every "continue development" and "takeover":

python scripts/security-scanner.py <project-directory> [--state-file .project-state.md]

---

Script Toolbox / 脚本工具箱

Script	Purpose	When to Use
project-scanner.py	Scan project structure, tech stack, components	Flow B (takeover), /rpd scan
security-scanner.py	Detect hardcoded secrets, injection, business security	Flow B, Flow C, /rpd scan
state-validator.py	Validate .project-state.md format and fields	After every state file change
gap-analyzer.py	Compare PRD features vs actual code implementation	Flow C (continue development)
state-guard.py	Backup and protect state file from corruption	Before updating state file
prd-validator.py	Check PRD completeness against template	After generating full PRD (A4)
run-eval.py	Run all evaluation scenarios	Testing RPD Skill itself

---

Failure Modes / 故障处理

EN	ZH	Handling
Empty project	空项目	Prompt: "Use 'new project' flow" / 提示"请使用「新建项目」流程"
Corrupted state file	状态文件损坏	Try to fix, else prompt to regenerate / 尝试修复，失败则提示重新生成
State version too high	状态文件版本过高	Prompt: "State file version incompatible, update RPD Skill"
Scan timeout	扫描超时	Degrade to manual tech stack questions / 降级为手动询问技术栈
Empty feature list	功能清单为空	Prompt: "Complete requirement alignment first" / 提示"请先完成需求对齐"

Warm Resume Phrases / 温暖续传话术

When resuming development, use warm, natural language instead of mechanical prompts:

恢复开发时，使用温暖自然的语言，而不是机械的提示：

EN examples:

• "Last time we were working on [project name], and we've completed [X] features. Want to continue? You can say 'continue' or tell me what you'd like to do next."
• "We've made good progress on [project name] — [X] features done. Ready to pick up where we left off?"

ZH examples:

• "上次我们在做 [项目名]，已经完成了 [X] 个功能。要不要继续？你可以说「接着来」或者告诉我下一步想做什么。"
• "[项目名] 进展不错——已完成 [X] 个功能。准备从上次停下的地方继续吗？"

---

Escalation / 熔断边界

EN	ZH	Handling
Concept PRD revised >3 times	概念版 PRD 修改 >3 次	Suggest user research first / 建议先做用户调研
Code contains hardcoded secrets	代码含硬编码密钥	Hard block, do not continue / 硬阻断，不继续
State file severely inconsistent with code	状态文件与代码严重不一致	Ask user which is correct / 提示用户确认
User rejects suggestions 3 times in a row	用户连续 3 次否定建议	Re-align requirements / 重新对齐需求