mb-map-codebase（mb 地图代码库）

mb-map-codebase — Brownfield: Repo → Memory Bank

• What it does: scans an existing repository, merges evidence from scoped workers, and writes as-is Memory Bank docs.
• Use it when: code already exists and you need a reliable baseline before planning changes.
• Input: repository root plus an initialized .memory-bank/.
• Output: product, architecture, testing, runbook, and contract docs that describe the current system without speculative backlog generation.

Preconditions

• You are in the repo root.
• .memory-bank/ skeleton exists (if not, run mb-init).

Process

1) Start a protocol + task folder

Create:

• .protocols/MAP-CODEBASE/plan.md
• .protocols/MAP-CODEBASE/progress.md
• .protocols/MAP-CODEBASE/verification.md (mapping completeness + evidence)
• .protocols/MAP-CODEBASE/handoff.md
• .protocols/MAP-CODEBASE/decision-log.md (optional)

Create:

• .tasks/TASK-MB-MAP/

Plan MUST include MB-SYNC (link .memory-bank/workflows/mb-sync.md).

2) Spawn scanning subagents (parallel)

Use ./agents/shared-repo-scanner.md as the base worker prompt.

Rules:

• Max 5–7 parallel subagents.
• Non-overlapping scopes (to avoid conflicts).
• Smart calling: each worker validates that its globs match real files.

Suggested scopes:

• S-01 tooling/CI
• S-02 backend/services
• S-03 frontend/UI
• S-04 data layer
• S-05 tests/quality

Each subagent writes:

• .tasks/TASK-MB-MAP/TASK-MB-MAP-S-0X-final-report-<code|docs>-01.md

3) Fan-in (mandatory)

Before writing final docs:

• read all worker reports
• merge into a single outline
• resolve contradictions (or record them as “needs verification”)
• write a short fan-in note into .protocols/MAP-CODEBASE/progress.md with:

- which reports were used - what conflicted and how resolved - open questions

4) Synthesize Memory Bank (as-is)

Using the fan-in view, create/update:

• product brief (.memory-bank/product.md)
• architecture overview (.memory-bank/architecture/)
• runbooks (.memory-bank/runbooks/)
• contracts (.memory-bank/contracts/)
• testing strategy (.memory-bank/testing/index.md)
• index (.memory-bank/index.md)

Use references/synthesis-checklist.md.

PRD-less rule (non-negotiable): if there is no prd.md, you MUST NOT create or populate: - .memory-bank/epics/* - .memory-bank/features/* - .memory-bank/tasks/backlog.md with waves/tasks Empty skeleton files/folders are OK if they were created by bootstrap. Mapping is documentation of the current system, not planning.

In overview docs, explicitly separate:

• Facts (evidence: paths/commands/logs/tests)
• Inferences (hypotheses, explicitly marked)

5) Ask for PRD delta

Once baseline MB exists:

• ask the user for a prd.md describing what to change/add.
• then run mb-from-prd to plan the delta on top of baseline.

6) Review gate

Run mb-review in fresh context.

7) MB-SYNC

Follow .memory-bank/workflows/mb-sync.md and append a changelog entry.

Definition of done

• .tasks/TASK-MB-MAP/ contains scoped reports.
• .memory-bank/ describes the system “as-is” with evidence.
• Contradictions and open questions are recorded.
• User has been asked for PRD delta.