ralph-o-matic
A job queue server that runs iterative AI coding refinement loops — locally via Ollama, against the Anthropic API, or through OpenRouter — so you can queue work, walk away, and review results as PRs.
The Problem
Iterating on code until tests pass and acceptance criteria are met produces excellent results, but doing it manually is tedious and doing it against cloud APIs burns credits fast. Most iterations are mechanical refinement, not creative problem-solving.
The Solution
Draft your implementation with Claude Code + Opus. Then hand off to ralph-o-matic (ROM), which runs the refinement loop with built-in circuit breakers, retry logic, session continuity, and per-iteration commits. Use local models via Ollama to save API credits, use the Anthropic API when you need Claude-grade quality, or use OpenRouter for access to models from multiple providers on a pay-per-token basis.
The Feature Pipeline
The /feature-pipeline skill (from the dbinky-skill-set plugin) automates the full lifecycle of building a feature — from idea to merged PR. You answer brainstorming questions, then walk away. Everything else runs unattended: design alignment, planning, implementation, refinement, and PR review. You wake up to a finished pull request.
Your Dev Env (Claude Code + Opus) ROM Server
┌─────────────────────────────────┐ ┌───────────────────────────┐
│ /feature-pipeline │ │ ralph-o-matic-server │
│ │ │ │
│ 1. Brainstorm product spec ◄─ you │ │
│ 2. Brainstorm impl. design ◄─ you │ │
│ ─── walk away ───────────────│──────── │ │
│ 3. Align designs to spec │ auto │ │
│ 4. Write implementation plans │ auto │ │
│ 5. Align plans to each other │ auto │ │
│ 6. Draft implementation │ auto │ │
│ 7. Generate ralph loop files │ auto │ │
│ 8. Submit to ROM ───────────────│───────────►│ Queue → Execute loop │
│ │ │ Commit → Push → PR │
└─────────────────────────────────┘ │ │
│ 9. Post-completion hook │
You get a Teams notification ◄────────────────│ → Auto PR review │
Review the PR, merge it │ → Teams notification │
└───────────────────────────┘
How it works
| Step | What happens | You involved? |
|---|
| 1 | Brainstorm product spec — interactive Q&A about what you want to build, outputs docs/specs/{slug}-spec.md | Yes |
| 2 | Brainstorm implementation design — interactive Q&A about how to build it, outputs docs/superpowers/specs/{slug}-design-phase-*.md | Yes |
| — | You walk away. Everything below is unattended. | — |
| 3 | Align designs to spec — reads spec and all design docs, fixes contradictions, commits | No |
| 4 | Write implementation plans — generates task files for each design phase | No |
| 5 | Align plans — cross-checks plans against spec/designs and each other, fixes gaps | No |
| 6 | Draft implementation — spawns parallel agents to execute plan tasks, runs tests, commits | No |
| 7 | Generate ralph loop files — auto-derives RALPH.md, focus-areas.md, gaps-identified.md from spec/design/plan artifacts | No |
| 8 | Submit to ROM — pre-flight checks, commit, push, submit job to ralph-o-matic server | No |
| 9 | Ralph refinement loop — ROM iterates: review a focus area, fix the most important issue, run tests, commit, repeat | No (server) |
| 10 | Post-completion hook — triggers automated PR review, applies fixes, sends Teams notification with PR link | No (server) |
Invoking the pipeline
/feature-pipeline Here's what I want to build: a user authentication system
with OAuth2, session management, and role-based access control.
It should support Google and GitHub as identity providers.
Flags:
--slug user-auth — override auto-derived feature slug--max-iterations 200 — override ralph iteration count (default: 200)--priority high — override ralph priority (default: high)--spec-only — stop after brainstorming (skip the automated pipeline)
Teams notifications
The pipeline sends Teams notifications at each milestone so you can monitor progress without watching a terminal:
