About
Code Guardian is a quality layer for AI-assisted software development. It runs inside AI coding assistants such as Claude Code or OpenAI Codex and checks their commands and file changes before they reach the software. It checks quality, security and software architecture through static code analysis, optimised skills and scripts, and it stops or reports whatever does not pass. Four locks halt risky operations before the command runs: deletions, secrets in a commit, deployments and database migrations. Before every change, the workflow has the assistant present a plan for a human to approve. Independent review agents read finished changes without knowing the plan and look for defects the writing agent missed. The package comprises five modes, seven gates, 15 hooks, 22 detectors, 27 skills and 14 review agents. Code Guardian grew out of more than 120 software projects and is in commercial use: currently version 17.10, with one edition for Claude Code and one for OpenAI Codex, running on macOS, Linux and Windows. Provimedia measures its effect in controlled comparisons of the same task with and without Code Guardian: in 10 valid comparisons the result was demonstrably better 5 times, undecidable 4 times and without a demonstrated difference once, and worse in none.