Superpowered Ralph: a phased Ralph Wiggum loop for Claude Code
Some days ago I was watching a video from Augusto Galego where he talks about Ralph Wiggum loops, a technique that comes from Geoffrey Huntley. The idea is you set up a loop where the AI keeps working through a task with clear phases instead of just going open-ended.
I was already using the Superpowers skill set for Claude Code a lot and my first thought was damn, imagine if I wired these two together. So I did.
I took the ralph-loop skill from Claude’s official repo and used Claude Code to help me wire it all up. It drives through design, planning, implementation, verification, review, and finish. Every feature starts with a failing test. If verification fails, it loops back with the failures documented. If code review finds real issues, it fixes them before moving on.
That last part is honestly what makes the biggest difference. Instead of spinning forever on the same problem, it knows when to simplify scope and try a different approach. That alone saved me a ton of time babysitting it.
Still rough around the edges. It runs as a skill so it manages context but doesn’t fully clear it between phases. But it persists state, tracks what it’s tried, and actually knows where it is in the process.
I ended up calling it Superpowered Ralph. It’s open source: github.com/jrlucas1/claude-superpowered-ralph