Skip to content

How To Use Claude Code with My BPE Workflow [Quickstart]

This quickstart guide gets you up and running with my personal Brainstorm-Plan-Execute (BPE) workflow for Claude Code. You'll install custom commands that structure your AI-assisted development into repeatable, reviewable steps.

Introduction

The BPE workflow breaks down AI-assisted development into three phases: Brainstorm your idea into a specification, Plan the implementation with generated prompts, and Execute each step with manual review. This approach produces maintainable code you actually understand.

For the full context behind this workflow, including Core Principles and detailed explanations, see How I Actually Use the Damn Thing.

Prerequisites

Before starting, ensure you have:

  • Claude Code installed via the native installer and either an Anthropic API key or a Claude Subscription
  • Ansible installed
  • Git installed
  • A code project to work on

Step 1 — Installing the BPE Workflow Commands

The BPE commands live in a homedir repository that uses Ansible for installation. Clone the repository and run the playbook:

cd ~
git clone https://github.com/MasonEgger/homedir.git
cd homedir
ansible-playbook ansible/setup.yml --tags claude

This installs:

  • ~/.claude/CLAUDE.md — Global configuration with development principles
  • ~/.claude/commands/ — Custom commands including the BPE workflow
  • ~/.claude/skills/ — Conditional context loading for languages like Python

Step 2 — Initializing Your Project

Navigate to your project directory and start Claude Code:

cd /path/to/your/project
claude

Before diving into work, set the output style to Explanatory mode. This makes CC explain what it's doing as it works, helping you learn and catch mistakes early:

/output-style

Select Explanatory from the menu. If you want even more detail (pair-programming style), choose Learning instead.

Now run /init to create your project context:

/init

This scans your project and creates a CLAUDE.md file with project-specific context. Review the generated file and commit it:

git add CLAUDE.md
git commit -m "Add project CLAUDE.md for AI context"
Prefer vim keybindings?

Run /vim to enable vim-style navigation and editing in Claude Code.

Step 3 — Brainstorming Your Feature

Start a brainstorming session by running the command with your idea:

/app-dev:brainstorm Build a CLI tool that converts PDF files to Markdown

Claude Code will ask you questions one at a time to refine your idea. Answer each question thoughtfully. When finished, it creates spec.md with your specification.

Review spec.md manually. Fix any errors or misunderstandings before proceeding. This file is the foundation for everything that follows.

Loading External Documentation

If your project uses a specific library or framework, you can load its documentation into context without copying it into your project. Clone the docs repo alongside your project:

cd ..
git clone https://github.com/some-library/docs.git some-library-docs
cd your-project

Then inside Claude Code:

/add-dir ../some-library-docs

CC can now reference those docs when answering questions or writing code that uses that library.

Step 4 — Creating a Plan

Clear your context and generate the implementation plan:

/clear
/app-dev:plan

This reads spec.md and creates two files:

  • plan.md — Step-by-step prompts for implementation, following Test-Driven Development
  • todo.md — Checklist mirroring the plan to track progress

Review plan.md manually. Verify the prompts make sense and follow your preferred patterns. Check that file paths are correct and test scenarios are appropriate.

Resuming a Session

Need to step away? You can close your terminal at any point. When you come back, start Claude Code and run:

/resume

This restores your previous session with full context intact.

Step 5 — Executing the Plan

Clear your context and start executing:

/clear
/app-dev:execute-plan

Claude Code reads todo.md to find the next unchecked item, loads the corresponding prompt from plan.md, and implements that step using TDD.

After each step completes:

  1. Review the generated code
  2. Manually run tests to verify they pass
  3. Commit the changes
  4. Run /clear to reset context
  5. Run /app-dev:execute-plan again for the next step

Repeat until todo.md is complete.

Conclusion

You now have a structured workflow for AI-assisted development:

  1. Brainstorm creates spec.md from your idea
  2. Plan generates plan.md and todo.md with implementation prompts
  3. Execute implements one step at a time with manual review between each

Saving Your Session

Before closing out, you can export the raw text of your session for archival or sharing:

/export

This saves the full conversation to a file you can reference later or share with teammates.

For the complete picture—including Core Principles like "This is MY Code, Not AI Code" and detailed explanations of Skills, context management, and more—read the full blog post: How I Actually Use the Damn Thing.