greenyogainc/varalign

VarAlign is a developer tool that tracks every variable an AI coding agent creates while writing code, remembers those variables across separate sessions, and checks them against what the code actually contains. Its main job is catching a common mistake AI assistants make: inventing a new name for something that already exists, like adding a variable called RETRY_LIMIT in one session when MAX_RETRIES already covers the same idea, or writing the same connection string under two different names across a project. The project has two parts under different licenses. The core engine, called varmem, is a command line tool with zero external dependencies that works with any editor or AI assistant, including Claude Code, Kilo, Cursor, GitHub Copilot, or code written by hand, since its detection is based on scanning files rather than plugging into a specific tool. A separate VS Code extension adds a visual interface on top, including tree views, an in-editor report, and jump to fix navigation. Under the hood, when an AI agent edits a file, a hook re-reads that file and records the variables it created, tagged with which session wrote them. Python files are parsed precisely using Python's own syntax tools, while other languages like JavaScript, PowerShell, and Bash use careful pattern based checks that account for each language's naming habits. All of this state is stored as plain, readable files inside a hidden folder in the repository itself, so it can be committed to version control and inspected with ordinary git tools. You can install it as a Claude Code plugin with a couple of commands, or set it up manually for Kilo Code or any local repository using the included command line script. It also includes a continuous integration mode that can fail a build pipeline when duplicate or misaligned variables are found above a chosen severity, and a local dashboard for reviewing issues across multiple registered repositories at once.