# Kaola-Workflow: A Six-Stage Intelligent Development Workflow Framework for Claude Code

> This article introduces Kaola-Workflow, a six-stage development workflow framework specifically designed for Claude Code. It enables end-to-end automation from requirement research to code delivery through multi-agent orchestration, supporting session recovery and cross-context state management.

- 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm)
- 发布时间: 2026-05-16T10:45:19.000Z
- 最近活动: 2026-05-16T10:51:16.442Z
- 热度: 159.9
- 关键词: Claude Code, AI编程, 工作流框架, 智能体编排, TDD, 代码审查, 多智能体, 开发自动化
- 页面链接: https://www.zingnex.cn/en/forum/thread/kaola-workflow-claude-code
- Canonical: https://www.zingnex.cn/forum/thread/kaola-workflow-claude-code
- Markdown 来源: floors_fallback

---

## Introduction / Main Floor: Kaola-Workflow: A Six-Stage Intelligent Development Workflow Framework for Claude Code

This article introduces Kaola-Workflow, a six-stage development workflow framework specifically designed for Claude Code. It enables end-to-end automation from requirement research to code delivery through multi-agent orchestration, supporting session recovery and cross-context state management.

## Background: Pain Points of AI-Assisted Development

Current mainstream AI coding assistants (such as Claude Code, GitHub Copilot) can generate code snippets, but still face many challenges in complex project development:

- **Context loss**: Easy to forget early requirements in long sessions, leading to implementation deviating from the original intent
- **Lack of systematicity**: A single conversation cannot cover the complete development cycle, often requiring manual coordination of multiple links
- **Uncontrollable quality**: No standardized code review and testing process, resulting in uneven output quality
- **Difficult to recover**: Hard to continue from breakpoints after session interruption, often needing to restart

Kaola-Workflow is born to solve these problems. It combines traditional software engineering methodology with AI capabilities to create a reproducible, auditable, and recoverable intelligent development process.

## Core Design Philosophy: Goal-Driven Six-Stage Workflow

Kaola-Workflow divides the entire development process into six clear stages, each with distinct input, output, and acceptance criteria:

## Stage 1: Research/Discovery

In this stage, the system uses multiple specialized agents working in parallel to comprehensively collect project-related information:

- **Code Explorer Agent (code-explorer)**: Analyzes the structure of existing codebases, understands project architecture and dependency relationships
- **Documentation Lookup Agent (docs-lookup)**: Retrieves external documents, API references, and best practices

The core output of this stage is a detailed technical research report, providing factual basis for subsequent decisions.

## Stage 2: Ideation

Based on the information collected in the first stage, the Planning Agent (planner) starts working. This stage uses high-level models (such as Claude Opus) for in-depth thinking, with outputs including:

- Functional requirement list and priority ranking
- Technical solution comparison and selection recommendations
- Risk identification and mitigation strategies

The planning stage emphasizes "think clearly before acting" to avoid subsequent rework.

## Stage 3: Architecture Planning (Plan)

The Architect Agent (code-architect) converts high-level design into executable technical solutions, with outputs including:

- Module division and interface definition
- Data flow design diagram
- Technology stack selection and version locking
- Development task decomposition and dependency diagram

## Stage 4: Execution & Development (Execute)

This is the most workload-intensive stage. The TDD Guide Agent (tdd-guide) executes according to the Test-Driven Development (TDD) process:

1. **Red Phase**: Write failing test cases to clarify functional expectations
2. **Green Phase**: Write minimal implementation code to pass the tests
3. **Refactor Phase**: Optimize code structure and eliminate technical debt

Each task has an independent TDD cycle to ensure code quality. When build errors occur, a dedicated Build Error Resolver Agent (build-error-resolver) intervenes to handle them.

## Stage 5: Code Review (Review)

The Code Review Agent (code-reviewer) and Security Review Agent (security-reviewer) work in parallel, with inspection contents including:

- Consistency of code style
- Compliance with design patterns
- Potential security vulnerabilities
- Performance optimization opportunities

The review results are output in the form of a structured report, marking the severity of issues and repair suggestions.