mb-from-prd
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
Chinesemb-from-prd — PRD → Memory Bank (greenfield)
mb-from-prd — PRD → 记忆库(全新项目)
- What it does: converts a PRD into structured project knowledge and traceable planning artifacts.
- Use it when: the project is greenfield and or equivalent requirements already exist.
prd.md - Input: or user-provided PRD text plus an initialized
prd.md..memory-bank/ - Output: product brief, RTM, epics, features, and concept docs ready for .
/prd-to-tasks
- 功能: 将PRD转换为结构化的项目知识和可追溯的规划工件。
- 适用场景: 项目为全新项目,且已存在或同等需求文档时使用。
prd.md - 输入: 或用户提供的PRD文本,以及已完成初始化的
prd.md目录。.memory-bank/ - 输出: 产品概要、RTM、Epic、Feature、概念文档,可直接用于流程。
/prd-to-tasks
Preconditions
前置条件
- You are in the repo root.
- exists (or the user provides PRD text).
prd.md - exists. If not, run
.memory-bank/first (or create the skeleton manually).mb-init
- 你当前处于代码仓库根目录。
- 存在文件(或用户提供PRD文本)。
prd.md - 存在目录。如果不存在,请先运行
.memory-bank/(或手动创建目录骨架)。mb-init
Process
流程
1) Load and sanity-check PRD
1) 加载PRD并进行合理性检查
- Read .
prd.md - Identify missing information and contradictions.
- Start a task protocol folder:
.protocols/PRD-BOOTSTRAP/- (steps)
plan.md - (Q/A and choices)
decision-log.md
- 读取。
prd.md - 识别缺失信息和矛盾内容。
- 创建任务协议目录:
.protocols/PRD-BOOTSTRAP/- (步骤说明)
plan.md - (问答记录和决策内容)
decision-log.md
2) Skills / tooling discovery (optional but recommended)
2) 技能/工具发现(可选但推荐)
If the PRD mentions “use skills / tools / CLIs”:
- run (project-installed first; marketplace second)
/find-skills - propose a minimal set of relevant skills to use (do not install without confirmation)
如果PRD中提到“使用技能/工具/CLI”:
- 运行(优先查找项目已安装的,其次查找市场中的)
/find-skills - 建议一组最小可用的相关技能(未经确认请勿安装)
3) Deep Questioning (rounds)
3) 深度提问(多轮)
Use .
./references/shared-deep-questioning.md- Ask questions in rounds of 3–5.
- After each round: summarize, update , and ask the next round.
decision-log.md - If user is temporarily unavailable: record in
Open questionsand stop. Do not proceed by inventing facts.decision-log.md
If the target mode is full autonomous:
- non-blocking gaps may be recorded as explicit
Assumptions - blocking gaps (security/compliance/external contracts/data-loss risks) must halt the run
使用。
./references/shared-deep-questioning.md- 每轮提出3-5个问题。
- 每轮结束后:总结内容,更新,然后提出下一轮问题。
decision-log.md - 如果用户暂时无法回复:在中记录
decision-log.md并停止流程,请勿自行编造信息继续推进。待解决问题
如果目标模式为完全自治:
- 非阻塞性的信息缺口可记录为明确的
假设条件 - 阻塞性缺口(安全/合规/外部合约/数据丢失风险)必须终止运行
4) Write L1 Product brief
4) 编写L1级产品概要
Update (use the user’s wording).
.memory-bank/product.md更新(使用用户的原始表述)。
.memory-bank/product.md5) Requirements + RTM
5) 需求 + RTM
Update :
.memory-bank/requirements.md- Enumerate REQ-IDs.
- Define “out of scope”.
- Fill RTM: REQ → Epic → Feature → Test.
更新:
.memory-bank/requirements.md- 枚举REQ-ID。
- 定义“超出范围”的内容。
- 填充RTM:需求ID → Epic → Feature → 测试。
6) Create Epics (L2, draft-first)
6) 创建Epic(L2级,优先草稿)
For each epic:
- Create
.memory-bank/epics/EP-<NNN>-<slug>.md - Use .
references/epic-template.md - Ensure business value + success metrics.
- Default until open questions are resolved.
status: draft
每个Epic对应:
- 创建
.memory-bank/epics/EP-<NNN>-<slug>.md - 使用模板。
references/epic-template.md - 确保包含业务价值 + 成功指标。
- 默认,直到所有待解决问题都被处理。
status: draft
7) Create Features (L3, draft-first)
7) 创建Feature(L3级,优先草稿)
For each feature:
- Create
.memory-bank/features/FT-<NNN>-<slug>.md - Use .
references/feature-template.md - Ensure autonomy and explicit acceptance criteria.
- Default until acceptance criteria + verification plan are solid.
status: draft
每个Feature对应:
- 创建
.memory-bank/features/FT-<NNN>-<slug>.md - 使用模板。
references/feature-template.md - 确保包含自治性和明确的验收标准。
- 默认,直到验收标准 + 验证计划都确认无误。
status: draft
8) Tasks planning (per-feature, no “everything at once”)
8) 任务规划(按Feature拆分,避免“一次性完成所有”)
Do not generate a full-task backlog “в лоб” for all features in one pass.
Instead:
- Seed with a short structure (waves + placeholders).
.memory-bank/tasks/backlog.md - For each selected feature, run to produce:
/prd-to-tasks FT-<NNN>.memory-bank/tasks/plans/IMPL-FT-<NNN>.md- atomic items grouped by waves
TASK-*
This keeps planning accurate and avoids speculative task explosions.
禁止一次性为所有Feature生成完整的任务待办列表。
正确做法:
- 为初始化简要结构(阶段 + 占位符)。
.memory-bank/tasks/backlog.md - 对每个选定的Feature,运行生成:
/prd-to-tasks FT-<NNN>.memory-bank/tasks/plans/IMPL-FT-<NNN>.md- 按阶段分组的原子化项
TASK-*
这种方式可以保证规划的准确性,避免产生大量推测性任务。
9) Duo docs for key concepts
9) 为核心概念创建双文档
For every concept that would otherwise require “reading many files” to understand later:
- (WHAT/WHY)
.memory-bank/architecture/<concept>.md - (HOW)
.memory-bank/guides/<concept>.md
对于后续需要“阅读大量文件”才能理解的每个概念:
- (是什么/为什么)
.memory-bank/architecture/<concept>.md - (怎么用)
.memory-bank/guides/<concept>.md
10) Update index
10) 更新索引
Update with annotated links to everything new.
.memory-bank/index.md在中添加所有新增内容的带注释链接。
.memory-bank/index.md11) Review gate
11) 评审关口
Run a fresh-context review (preferably ).
mb-review运行一次新上下文评审(优先使用)。
mb-review12) Autonomous handoff (optional)
12) 自治交接(可选)
If the goal is “PRD → done without more user interaction”:
- do not execute tasks from here manually
- hand off to generated project command
/autonomous
如果目标是“无需更多用户交互,从PRD到交付完成”:
- 请勿从此处手动执行任务
- 交接给生成的项目命令
/autonomous
Definition of done
完成标准
- product.md + requirements.md are coherent.
- Every REQ maps to an Epic/Feature in RTM.
- Epics and features exist with acceptance criteria.
- backlog.md exists as a plan skeleton; feature-level tasks are produced via .
/prd-to-tasks - index.md is updated.
- product.md + requirements.md内容连贯一致。
- 每个需求ID都在RTM中映射到对应的Epic/Feature。
- Epic和Feature都已创建,且包含验收标准。
- 存在作为规划骨架的backlog.md;Feature级任务通过生成。
/prd-to-tasks - index.md已更新。