brajesh2022/vcs-edit-tools

VCS Edit Tools is an experimental command line tool aimed at making AI coding assistants cheaper to run. The problem it targets is output tokens, the words an AI writes back out, which are often billed at a higher rate than the words it reads in. Most coding tools ask an AI to make an edit by having it type out the old block of code it wants to replace, followed by the new block, so the tool can match the old text and swap it. That approach is safe because it matches on actual content, but it means the AI spends output tokens retyping code it already just read, before it even gets to the new code that matters. This project's fix is to have its read command hand back a file as numbered lines along with an identifier for that exact snapshot, called a blob. When the AI later wants to make an edit, instead of retyping the old block, it just supplies the blob identifier, the line range from that snapshot, and the new content. The tool finds that range inside the stored snapshot rather than the live file, applies the edit there, and then merges the result into whatever the file currently looks like using git as the merge engine. Because the anchor is the frozen snapshot rather than live line numbers, the tool avoids the usual problem where line numbers drift out of date after earlier edits. The project includes several commands beyond read and replace, including one that lists a directory more lightly than typical tools, one that returns a file's structural skeleton before reading it in full, and one that batches multiple edits into a single call. Installation adds instructions to whichever AI tool you use, whether that means a rules file, an agents file, or specific plugins and hooks depending on the tool. The author is upfront that this only works reliably on small files right now, since larger file content gets cut off at a 16 kilobyte limit in many command output channels, so it is described as a proof of concept rather than a finished, general solution.