karurikwao/signaltrim

SignalTrim is a tool that makes AI coding assistants, such as Claude Code, Codex, Gemini CLI, Cursor, and Windsurf, give shorter and less padded answers while keeping the technical content intact. Instead of long explanations full of filler phrases, it trims replies down to the facts: the fix, the command, the error, or the code, without the extra wrapping language many AI assistants tend to add. It installs as a local skill, hook system, and command set across more than 30 supported coding agents, using a single install script that works on macOS, Linux, Windows, and WSL. Once installed, a user turns it on with a command like /signaltrim or by saying use SignalTrim mode, and turns it back off by asking for normal mode. Several modes are available, ranging from full sentences with no hedging, to fragments with articles dropped, to a maximum terseness setting, plus lighter and heavier compression styles written in a classical Chinese register for users who want that. The tool keeps a reply in the user's original language: text in Portuguese, Spanish, or French stays in that language even as it gets shorter. Beyond compressing replies, SignalTrim adds extra commands for writing commit messages in a short, useful format, producing dense notes for pull request reviews, and reading a user's local session logs to estimate how many output tokens were saved. It runs entirely on the user's own machine: nothing is sent to an outside server, and the project states plainly that it does not reduce the tokens an assistant reads as input, only the length of what it writes back. The project is written in JavaScript, released under the MIT license, and requires Node.js version 20.19 or later. It ships with test and verification commands so a user or maintainer can check that the install and the compression logic work correctly before relying on them. With only a handful of stars so far, this looks like a newer, early stage project.