gitmyhub

code-is-cheap

Shell ★ 55 updated 17d ago

A staged, human gated workflow framework of skills for running AI coding agents like Claude Code through a controlled review process.

ShellPythonClaude CodeCodextmuxsetup: moderatecomplexity 4/5

code-is-cheap is an engineering framework of skills and scripts for Codex and Claude Code that turns AI assisted coding into a controlled, staged process rather than a loose set of prompts. The core idea is that a human first prepares the architecture, phase boundaries, and an implementation control plan, and after that a controller agent can automatically advance the work through each phase until a local branch is ready to open as a draft pull request.

Rather than letting an AI agent freely invent design decisions as it goes, the framework walks through a fixed sequence: reviewing a plan, implementing it in small pieces, reviewing the code, fixing any issues found, reviewing again, running a deeper aggregate review at the end, and tracking any remaining risks. The controller only stops and asks a human for input when there is a genuinely blocking question, unclear ownership, a failed validation step, a risk that needs human judgment, or when it is about to take an external action like opening a pull request or merging code. Once a person approves opening the draft pull request, the framework can continue automatically through pull request review, fixes, and further pushes.

The repository ships five named skills. Gateflow advances a feature from planning through to an approved draft pull request. Phaseflow works from a design document and an implementation control document to plan and track work phase by phase. Planreview gives an adversarial review of a plan before implementation starts. Deepreview performs a strict code review of a workspace, pull request, or whole repository. Init agents helps a controller agent coordinate with other AI agents running in separate terminal panes using tmux.

To use it, you need Codex, Claude Code, or a similar tool that supports local skill style instruction files, and Python 3.11 or newer if you want to run the included skill validator. Setting up multi agent coordination additionally requires tmux and a companion tool called tmux-cli. The full README is longer than what was shown.

Where it fits