# Arcgentic

A harness engineering layer around Codex or Claude Code that turns ad-hoc prompting into a gated workflow of planning, building, self-audit, optional user testing and independent audit before a task can close.

Arcgentic is a harness engineering layer for AI coding agents, giving Codex and Claude Code roles, handoffs, stop states and audit gates around the model rather than adding a new coding agent of its own.

A task moves through clarifying the idea, planning the work, building it, a self-audit, an optional realistic user test, and an independent external audit, closing only when the evidence is good enough, rather than stopping because the assistant said it was done. It has a configurable role-routing topology engine, an MCP-UI live status panel, and native dispatch support for Claude Code V2.

It packages the development discipline extracted from a real agent project that survived more than 30 strict development rounds, and is distributed as a plugin for Claude Code, Codex, OpenClaw and through PyPI.

Website: https://github.com/Arch1eSUN/Arcgentic