Analysis updated 2026-05-18
Turn a rough project idea into a description, user stories, database schema, and a numbered build plan.
Convert a feature request into a formal specification and phased plan with human approval checkpoints.
Run an AI coding agent through a phased plan automatically, committing each completed phase to git.
Keep documentation of an existing codebase's architecture and conventions in sync with the actual code.
| beerandcodeteam/beer-and-code-harness | echo-of-machines/fable-advisor | roland-luo/codex-dream-skin | |
|---|---|---|---|
| Stars | 26 | 26 | 26 |
| Language | Shell | Shell | Shell |
| Setup difficulty | moderate | easy | easy |
| Complexity | 3/5 | 2/5 | 2/5 |
| Audience | developer | developer | general |
Figures from each repo's GitHub metadata at analysis time.
Requires Claude Code or OpenAI's Codex CLI with an API key, and a clean git repository to run ralph.sh.
Beer and Code Harness, also called bc-harness, is an add-on plugin for Claude Code, an AI coding assistant made by Anthropic, that helps take a software project from a rough idea all the way to working code in a structured way. Instead of asking the AI to write code in one big undirected pass, it breaks the process into three separate pipelines that connect together. The first pipeline, started with the init command, walks you through describing your project, then generates user stories, a database design, and a numbered plan of build phases, asking questions along the way. The second pipeline, plan, takes a description of a specific feature and turns it into a formal specification document, an architecture aware task breakdown, and a phase by phase plan, with checkpoints where a human developer reviews and approves each step before moving on. A third command, ai-context, reads the code that has already been written and keeps a set of reference documents about it up to date, so the other two pipelines always have accurate information about the current state of the project. Once a plan of phases exists, a separate script called ralph.sh can run through it on its own. It reads each phase from the plan document and hands it to a brand new AI coding session, one phase at a time, with no person needed in between, and it commits the finished work to version control after each phase passes its checks. The tool itself does not dictate which programming language, framework, or coding conventions a project uses. Those come from the project's own documentation files, so it can be used with any tech stack. Running ralph.sh requires either OpenAI's Codex command line tool with an OpenAI API key, or Anthropic's Claude Code with an Anthropic API key, and it must be run from the root of a git repository whose working tree has no uncommitted changes.
A Claude Code plugin that turns a project idea into a phased build plan, then runs an AI agent through each phase automatically with human checkpoints.
Mainly Shell. The stack also includes Shell, Claude Code, Codex CLI.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.