mb-from-prd（mb 来自 珠三角）

mb-from-prd — PRD → Memory Bank (greenfield)

• What it does: converts a PRD into structured project knowledge and traceable planning artifacts.
• Use it when: the project is greenfield and prd.md or equivalent requirements already exist.
• Input: prd.md or user-provided PRD text plus an initialized .memory-bank/.
• Output: product brief, RTM, epics, features, and concept docs ready for /prd-to-tasks.

Preconditions

• You are in the repo root.
• prd.md exists (or the user provides PRD text).
• .memory-bank/ exists. If not, run mb-init first (or create the skeleton manually).

Process

1) Load and sanity-check PRD

1. Read prd.md.
2. Identify missing information and contradictions.
3. Start a task protocol folder:

- .protocols/PRD-BOOTSTRAP/ - plan.md (steps) - decision-log.md (Q/A and choices)

2) Skills / tooling discovery (optional but recommended)

If the PRD mentions “use skills / tools / CLIs”:

• run /find-skills (project-installed first; marketplace second)
• propose a minimal set of relevant skills to use (do not install without confirmation)

3) Deep Questioning (rounds)

Use ./references/shared-deep-questioning.md.

• Ask questions in rounds of 3–5.
• After each round: summarize, update decision-log.md, and ask the next round.
• If user is temporarily unavailable: record Open questions in decision-log.md and stop. Do not proceed by inventing facts.

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

4) Write L1 Product brief

Update .memory-bank/product.md (use the user’s wording).

5) Requirements + RTM

Update .memory-bank/requirements.md:

• Enumerate REQ-IDs.
• Define “out of scope”.
• Fill RTM: REQ → Epic → Feature → Test.

6) Create Epics (L2, draft-first)

For each epic:

• Create .memory-bank/epics/EP-<NNN>-<slug>.md
• Use references/epic-template.md.
• Ensure business value + success metrics.
• Default status: draft until open questions are resolved.

7) Create Features (L3, draft-first)

For each feature:

• Create .memory-bank/features/FT-<NNN>-<slug>.md
• Use references/feature-template.md.
• Ensure autonomy and explicit acceptance criteria.
• Default status: draft until acceptance criteria + verification plan are solid.

8) Tasks planning (per-feature, no “everything at once”)

Do not generate a full-task backlog “в лоб” for all features in one pass.

Instead:

1. Seed .memory-bank/tasks/backlog.md with a short structure (waves + placeholders).
2. For each selected feature, run /prd-to-tasks FT-<NNN> to produce:

- .memory-bank/tasks/plans/IMPL-FT-<NNN>.md - atomic TASK-* items grouped by waves

This keeps planning accurate and avoids speculative task explosions.

9) Duo docs for key concepts

For every concept that would otherwise require “reading many files” to understand later:

• .memory-bank/architecture/<concept>.md (WHAT/WHY)
• .memory-bank/guides/<concept>.md (HOW)

10) Update index

Update .memory-bank/index.md with annotated links to everything new.

11) Review gate

Run a fresh-context review (preferably mb-review).

12) Autonomous handoff (optional)

If the goal is “PRD → done without more user interaction”:

• do not execute tasks from here manually
• hand off to generated project command /autonomous

Definition of done

• 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.