wtf is claude-squelch?

Squelch is a small tool that trims down noisy command output before it reaches Claude Code, an AI coding assistant. When you run something like a test suite or a package install, the terminal often prints thousands of lines, most of which are just passing tests or progress messages. Squelch catches that output and shrinks it down to the useful parts before the AI ever sees it, which saves a large amount of the context space the AI has available to think. It works as a hook that plugs into Claude Code's own hook system. When a command like pytest or npm install finishes running, squelch looks at what kind of command it was and, if it recognizes the pattern, rewrites the output to keep failures, warnings, and summaries while dropping the repetitive lines that show everything passed. According to the project's own benchmarks, this can cut token usage for noisy commands by 70 to 95 percent, while commands whose output is already mostly useful, like a linter or type checker, are left untouched. Safety is a core design goal here. The tool is built to fail toward showing too much rather than too little: errors and warnings are always kept, any command it does not recognize passes through completely unchanged, and if anything goes wrong internally it simply does nothing rather than risk hiding a real problem. Anyone can also bypass it entirely for a single command if they want the raw output. Installation takes about thirty seconds, either through Claude Code's plugin marketplace or by manually adding a couple of lines to a settings file that tells Claude Code to run the squelch script after every terminal command. The tool covers common test runners, package managers, build tools, linters, and git commands, and is written in Python with no external dependencies. It comes from a larger collection of reusable Claude Code hooks the same author maintains, and is released under the MIT license.