Super-Ralph
Ralph's autonomous development loop + Superpowers' disciplined methodology = production-ready autonomous AI development.
Super-Ralph fuses two proven open-source projects into a unified system:
- Ralph (MIT) -- An autonomous AI development loop for Claude Code with intelligent exit detection, rate limiting, circuit breakers, session continuity, and a comprehensive bash-based orchestration system (465+ tests, 100% pass rate).
- Superpowers (MIT) -- A composable skills framework and software development methodology that enforces brainstorming, TDD, systematic debugging, two-stage code review, and verification-before-completion workflows across AI coding agents.
Ralph gives your AI agent endurance (it can run autonomously for hours). Superpowers gives it discipline (it follows engineering best practices instead of ad-hoc implementation). Super-Ralph combines both.
Table of Contents
Installation (Claude Code)
Note: Super-Ralph includes all skills from the Superpowers plugin. If you have Superpowers installed, you can uninstall it after installing Super-Ralph to avoid duplicate skills.
Step 1: Add the marketplace
In Claude Code, run:
/plugin add https://github.com/aezizhu/super-ralph
This registers Super-Ralph as a plugin marketplace. Claude Code will clone the repo and discover the plugin.
Step 2: Install the plugin
After adding the marketplace, Claude Code will show super-ralph as an available plugin. Select it to install.
Step 3: Restart Claude Code
Exit and reopen Claude Code. Start with:
/using-super-ralph "Your task here" --max-iterations 15
This starts the autonomous loop with full methodology enforcement. All 14 skills + 3 commands are available:
# Commands
/using-super-ralph Start Super-Ralph loop (main entry point)
/sr-cancel-ralph Cancel an active loop
/sr-help Show all commands and skills
# Skills (auto-triggered)
/sr-brainstorming Explore intent, requirements, and design before coding
/sr-writing-plans Create detailed implementation plans with bite-sized tasks
/sr-test-driven-development Strict RED-GREEN-REFACTOR TDD workflow
/sr-systematic-debugging 4-phase root cause investigation before any fix
/sr-verification-before-completion Evidence before claims — run commands, read output
/sr-subagent-driven-development Fresh subagent per task with two-stage review
/sr-executing-plans Execute plans in batches with human checkpoints
/sr-requesting-code-review Dispatch code-reviewer subagent on git diff
/sr-receiving-code-review Technical evaluation of review feedback, not blind agreement
/sr-finishing-a-development-branch Verify tests → merge/PR/keep/discard
/sr-using-git-worktrees Isolated workspaces for feature development
/sr-dispatching-parallel-agents One agent per independent problem domain
/sr-writing-skills TDD for skills — test scenarios before writing
Updating
/plugin update super-ralph
Uninstalling
Use /plugin to browse installed plugins and remove Super-Ralph.
Why Super-Ralph Exists
Ralph alone is a powerful autonomous loop, but it has gaps that Superpowers fills:
| Ralph Default | Super-Ralph Enhancement |
|---|
| Jumps straight to implementation | Brainstorms design first, validates with user |
| Tests as afterthought | Enforces strict TDD: test first, watch fail, implement |
| Works on main branch | Uses git worktrees for isolation |
| Reviews code informally | Two-stage review: spec compliance then code quality |
| "Fix plan says done" = done | Verification-before-completion: evidence before claims |
| Serial task execution | Subagent-driven development with fresh agents per task |
| Guesses at bug fixes | Systematic 4-phase debugging: root cause before fix |
| No structured planning | Bite-sized tasks with exact file paths and commands |
What's Inside
