# RunWield

An AI coding harness that has an agent write a plan for a code change and wait for the developer's approval before touching any files, then verifies the finished work against CI and a separate reviewer before calling it done.

RunWield reads a repository, writes a plan for a requested change, and only lets specialized agents build it once the plan is approved, rather than starting to edit code immediately; it will not mark work complete until continuous integration and a separate reviewer confirm the result matches the approved plan. Every decision made along the way is recorded, so a later change in the same project can draw on what earlier sessions already learned instead of starting from zero each time.

The amount of process scales with the risk of the change: a quick fix stays quick, while larger or riskier work gets an expert architect review and is split into a series of plans implemented in smaller, manageable chunks. It works with any model provider, either through a subscription login or an API key, and runs locally on macOS and Linux, installing to a user's home directory with no root access required.

The core product is free, and the project is distributed as a shell installer script.

Website: https://runwield.dev