A quality layer inside your AI coding assistant
Code Guardian is a quality layer for AI-assisted software development. It runs inside an AI coding assistant, either Claude Code or OpenAI Codex, rather than alongside it, and checks the assistant's commands and file changes before they reach your codebase. It covers quality, security and software architecture, using static code analysis together with optimised skills and scripts, and it stops or reports anything that fails. Code Guardian grew out of more than 120 software projects and over 20 years of development experience.
The current version is 17.16. It comes in two editions, one for Claude Code and one for OpenAI Codex; both carry the same version number and are built from the same source. The Claude Code edition includes five modes, seven gates, 15 hooks, 22 detectors, 31 skills and 14 review agents.
What it checks and what it stops
Four safeguards stop risky operations before they run: deletions, commits containing secrets, deployments and database migrations. Rather than flagging the problem in a report afterwards, Code Guardian refuses commands that would destroy work. In Codex, the safeguards are only armed once you have confirmed them, and that confirmation has to be repeated after every update.
In Claude Code, the workflow tells the agent to work in plan mode and submit its plan for human approval before it changes any files. This is a workflow instruction rather than a hard lock. Any outstanding decisions are put to you one at a time, each with a recommendation up front. In Codex, Code Guardian recommends planning before the first file is changed.
Once the work is done, independent review agents read the finished changes cold: without the conversation, without the plan and without the author's own assessment. That way they catch defects the coding agent missed.
The 31 skills also cover the work around the code. They include a task list and task folder that preserve what a long conversation would otherwise forget, a root-cause analysis that separates cause from symptom, a full-repository audit for AI slop, and tools that set up static analysis for PHP, JavaScript/TypeScript/Vue and Python projects.
What you need and how it is licensed
Code Guardian requires Python 3.9 or newer and either Claude Code or OpenAI Codex, and runs on macOS, Linux and Windows. On Windows it works natively through Git for Windows, so WSL is not needed.
You can work with Code Guardian in English; only its predefined hook messages, guides and update notes are in German. The product page is available in English and German.
It is sold as a one-off company licence with a perpetual right of use, covering any number of developers within the licensed company. Updates are available through an optional monthly subscription that can be cancelled at any time. Code Guardian is developed by Provimedia GmbH in Bietigheim-Bissingen, Germany.
