mattiaf95/codebase-analysis-ai

Codebase Analysis AI is a tool that keeps a project's documentation in sync with its actual code, so a new developer can read the docs and understand the project right after cloning it. Instead of scanning an entire repository every time, it starts from what changed in Git, figures out which documentation files are affected by those changes, and asks an AI agent to update only the parts that need a human style judgment call. The system has a Python checker that maps source files to the documents they relate to, and it stops after checking one level of related documents rather than following every link in the whole documentation set. It uses content hashes to track which files have already been reviewed, even if the changes were not yet committed or were rebased later. For a brand new project with no documentation yet, there is a full setup mode that first builds an inventory of the codebase's structure, then groups the code into logical areas and matching documentation topics. For larger areas of code, it can create separate, read only review agents that look only at their assigned area and report their findings as structured data, which a main agent then uses to write the actual documentation. The tool supports several modes: setting itself up, building documentation from scratch, keeping documentation updated as code changes, auditing documentation for accuracy without changing anything, restructuring existing documentation, and a lightweight check mode that runs without needing an AI model at all, useful for automated checks before a commit or in a continuous integration pipeline. It is designed to work with more than one AI coding assistant, including Claude Code, OpenAI Codex, Gemini CLI, and GitHub Copilot, each with its own way of activating the skill and its own location for the generated review agent profiles. If documentation already exists in a project, the tool always asks the user whether to archive the old version or replace it, rather than silently rewriting things.