# Be-My-Butler: A Multi-Agent Validation-Driven Workflow Orchestration Framework for Claude Code

> BMB is a multi-agent orchestration framework designed for Claude Code CLI. It addresses common issues of AI programming assistants such as hallucinations, omissions, and self-review biases through a 12-step pipeline, cross-model blind review validation, and a three-layer compression protocol.

- 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm)
- 发布时间: 2026-04-05T22:15:23.000Z
- 最近活动: 2026-04-05T22:18:42.580Z
- 热度: 141.9
- 关键词: AI编程, Claude Code, 多智能体, 代码审查, 跨模型验证, 智能体编排, 软件工程, 自动化工作流
- 页面链接: https://www.zingnex.cn/en/forum/thread/be-my-butler-claude-code
- Canonical: https://www.zingnex.cn/forum/thread/be-my-butler-claude-code
- Markdown 来源: floors_fallback

---

## Introduction: Core Overview of the Be-My-Butler Framework

Be-My-Butler (BMB for short) is a multi-agent orchestration framework designed for Claude Code CLI, aiming to solve common issues of AI programming assistants such as hallucinations, missing boundary conditions, and self-review biases. Its core enhances code reliability through a 12-step pipeline, cross-model blind review validation, and a three-layer compression protocol, adhering to the "slow is fast" philosophy and suitable for production environments requiring high-quality code.

## Background: The Reliability Dilemma of AI Programming

With the popularization of AI programming assistants, developers have found that existing tools (such as Cursor, GitHub Copilot) generate code quickly but have issues like hallucinations, missing boundary conditions, and self-review biases. When the same model both writes and reviews code, it's hard to detect its own errors. The BMB project addresses this pain point by improving code reliability through multi-agent collaboration and cross-model validation mechanisms.

## Methodology: Core Mechanisms and Architecture Design

BMB is designed around five key issues: 1. Cross-model blind review to solve self-review bias; 2. Council Debate mechanism to address narrow design vision; 3. Three-layer compression protocol (intra-step, inter-step, session-level) to control context explosion; 4. Divergent Framing technology to identify assumption loopholes; 5. FTS5 knowledge base to solve knowledge loss. The core is a 12-step pipeline architecture covering session preparation, brainstorming, debate and decision-making, architecture design, execution, testing, validation, etc., and defines ten professional agent roles (Lead, Consultant, Architect, etc.) for collaborative work.

## Practice: Adaptive Workflow and Knowledge Management

BMB provides a "recipe" system to adapt to different tasks: feature (full 12 steps), bugfix (skip brainstorming/debate), refactor (skip frontend), etc. Knowledge management uses a three-layer system: project local, global cross-project, and CLAUDE.md solidified rules. Full-text search is implemented via SQLite FTS5, and repeated issues are automatically promoted to candidate rules.

## Recommendations: Deployment and Usage Guide

BMB deployment depends on Claude Code CLI, tmux, python3, sqlite3, git. Optional Codex/Gemini CLI can be used to enable cross-model validation. After installation, use 'bmb doctor' to verify dependencies. Practical recommendations: Start configuration from /BMB-setup, first try the bugfix recipe to familiarize with the process, then gradually enable complex recipes and cross-model validation.

## Conclusion: A Reliability-First AI Programming Paradigm

BMB represents a reliability-first AI programming philosophy. Instead of pursuing speed, it improves code quality through multi-agent collaboration, structured processes, and cross-model validation. For teams pursuing high quality, its design ideas and mechanisms are worth studying and can provide references for AI-assisted development practices.