mcpField
M
mb-review

mb-review

github.com/mrvladd-dTesting & quality5640 installsUpdated 21w ago
Repository

About mb-review

mb-review is a skill that coordinates independent reviews of an existing Memory Bank across architecture, scope, planning, security, and compliance. It synthesizes findings into prioritized fixes and a verdict before execution or after project setup.

Key features

  1. 01Run independent reviews in fresh contexts
  2. 02Review architecture, scope, plans, and security
  3. 03Check Memory Bank discipline and traceability
  4. 04Prioritize findings and synthesize fixes
  5. 05Set an approval gate for autonomous execution

Use cases

  1. 01Review Memory Bank after project bootstrap
  2. 02Find documentation gaps before task execution
  3. 03Gate autonomous work on unresolved review findings

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/mrvladd-d/memobank --skill mb-review

After installing, type/mb-reviewin Claude Code to confirm it loaded.

Details

From the project README

mb-review — Multi-expert Memory Bank review

  • What it does: runs independent reviewers over architecture, scope, task planning, security, and Memory Bank discipline.
  • Use it when: you need a clean review gate before execution, after bootstrap, or before trusting the docs.
  • Input: an existing .memory-bank/.
  • Output: reviewer reports in .tasks/TASK-MB-REVIEW/ plus a synthesized list of fixes and review verdicts.

Goal

Detect gaps, contradictions, broken traceability, and non-compliance early.

This is not the same as per-task semantic verification. Use mb-red-verify when the question is “did this task solve the right problem in substance?”

Preconditions

  • .memory-bank/ exists.

Process

1) Create review task folder

Create:

  • .tasks/TASK-MB-REVIEW/
2) Spawn reviewers (fresh contexts)

Spawn these subagents in parallel (max 5–7):

  1. Architect — ./agents/shared-review-architect.md
  2. Scope/RTM — ./agents/shared-review-scope.md
  3. Task planning — ./agents/shared-review-plan.md
  4. Security — ./agents/shared-review-security.md
  5. MBB compliance — ./agents/shared-mb-reviewer.md
  6. Code quality (conditional: if repo has code) — ./agents/shared-review-code.md

Each reviewer must:

  • write a detailed report to .tasks/TASK-MB-REVIEW/
  • return only a short summary + verdict to the orchestrator
3) Synthesize and decide

As orchestrator:

  • combine findings
  • deduplicate
  • rank issues P0–P3
  • produce a concrete fix plan

If the repo is preparing for /autonomous:

  • treat unresolved P0/P1 issues as blocking
  • do not allow batch execution until the final verdict is APPROVE
4) Gate

If any reviewer returns REJECT:

  • fix MB
  • re-run mb-review

Definition of done

  • .tasks/TASK-MB-REVIEW/ contains the reviewer reports.
  • Orchestrator produced an actionable prioritized fix list.
  • Final verdict: APPROVE.

Comments

Markdown supported