rea-dev 0.7.0

Development toolkit — bootstraps structured Claude Code workflow (commands, agents, plans, CI) into any project

REA

Structured development workflow for Claude Code.

REA installs slash commands, composable agents, and CI config into any project. The CLI copies files — all intelligence runs through Claude.

pip install rea-dev
rea setup .
# open Claude Code → /rea-init

What you get

Slash commands (use inside Claude Code):

Command	What it does
/rea-plan	Write a spec + todo, adversarial review, get approval before coding
/rea-execute	Agents implement in parallel with spec review + code review
/rea-commit	Commit, push, open PR to the right branch automatically
/rea-brainstorm	Explore ideas before committing to a plan
/rea-verify	Health check — CI, branches, config
/rea-write-skill	Create a new custom agent or command
/rea-wrap	End-of-session summary + save lessons
/rea-update	Update REA + sync templates

Composable agents (called by commands or standalone):

Agent	Purpose
explorer	Read-only codebase research
implementer	TDD-driven implementation
spec-reviewer	Verifies implementation matches requirements
code-reviewer	Code quality assessment
debugger	4-phase root cause debugging
plan-reviewer	Adversarial plan review
dispatcher	Groups todo items into parallel/sequential batches
bug-scanner	Logic bugs, edge cases, error handling gaps
security-scanner	Security vulnerabilities
skill-writer	Creates new agents/commands matching conventions

Quick start

Requirements: Python 3.11+, gh CLI authenticated, git repo with GitHub remote.

# Install
pip install rea-dev

# Add to your project
rea setup /path/to/project

# Finish setup in Claude Code
/rea-init

/rea-init detects your stack and installs CI workflows, branch protection, hooks, and a project-specific CLAUDE.md.

How it works

1. Plan — /rea-plan researches your codebase, drafts a spec + todo, runs adversarial review, and waits for your approval before any code is written.

2. Execute — /rea-execute dispatches todo items to parallel agents. Each item goes through implementation, spec review, and code review.

3. Commit — /rea-commit detects your branch, commits, pushes, and opens a PR to the right target (feature/* → staging, staging → main).

Plans persist in .rea/plans/ across sessions. The NEXT: marker tracks where you left off.

Branch strategy

feature/*  →  PR to staging
staging    →  PR to main (release)
hotfix/*   →  PR to main (emergency)

Architecture

Commands are orchestrators. Agents are building blocks. Agents never call other agents — only commands orchestrate.

License

MIT