pegasusdev18/git-commit-craft

git-commit-craft is a command line tool built with Node.js that uses Google's Gemini AI model to handle two common but tedious developer tasks: writing well-formed commit messages and drafting README files. Instead of staring at a blank prompt trying to summarize a code change, or writing project documentation from scratch, a developer can run one of two commands and let the AI generate a draft based on the actual project content. For commit messages, the tool reads the output of git diff, either for changes already staged or, if nothing is staged, for unstaged changes instead. It sends that diff to the Gemini model along with instructions to follow the Conventional Commits format, a widely used standard for structuring commit messages. The AI's suggested message is shown in the terminal, and the developer can accept it as is, edit it, ask for a new version, or cancel entirely before anything is committed to the repository. For README generation, the tool scans the project's folder structure, figures out which programming language is used most based on file extensions, and reads details from the package.json file if one exists, such as the project name, dependencies, and available scripts. It sends all of that information to the same AI model with instructions to write a complete, professional README, then saves the result directly to a README.md file, asking first if one already exists. Using the tool requires a free Gemini API key from Google AI Studio, which can either be typed in interactively the first time a command runs, where it gets saved locally on your computer, or set as an environment variable. The README notes that Google now requires these API keys to be restricted to a specific API address, and that keys created directly through Google AI Studio already meet that requirement automatically. The project is installed by cloning the repository, running npm install, and linking it so the command becomes available from any git repository on the same machine, and it is released under the MIT license.