I built SimpleRalph — a file-driven autonomous coding loop for any repo

I just open-sourced SimpleRalph:

https://github.com/Seungwan98/SimpleRalph

SimpleRalph is a lightweight CLI for running a file-driven autonomous coding loop inside any repository.

The core idea is simple:

• give it one topic
• create a local session under .simpleralph/
• keep the loop state explicit with PRD / Tasks / Status / Log
• run compile/test gates between iterations
• export artifacts when needed

Why I made it

A lot of agent workflows feel either:

1. too chat-memory-driven to resume cleanly
2. too opaque to inspect when something goes wrong

I wanted something more explicit and reviewable.

SimpleRalph keeps the working state on disk instead of burying everything in one long conversation.

Current commands

• simpleralph init
• simpleralph run
• simpleralph status
• simpleralph export

A small detail I cared about

If the target repository contains an AGENTS.md, the default session config makes that file available to the agent run automatically.

The package is agent-agnostic at the config level, even though the default generated command template is OpenCode-friendly.

Status

It is still alpha, but the current GitHub version has already been tested from a fresh clone with:

• install
• init
• status
• export
• one safe loop run

If you try it, I would love feedback on:

• whether the file-driven model feels useful
• where the UX feels too heavy
• which agent CLIs should be better supported first