What happens when you test a plugin designed to make Claude Code responses shorter against simply typing "be brief"?
Developer Max Taylor ran that comparison, pitting the Caveman plugin against a two-word system prompt instruction. The Caveman plugin exists because Claude Code's default responses run long - it explains its reasoning, lists alternatives, and adds context you don't need when moving fast through a codebase. The plugin handles brevity automatically so you don't have to ask for it in every session.
The benchmark questions whether that dedicated tooling is necessary. Plugins carry overhead: they need updating, can conflict with other extensions, and add complexity to a dev environment already juggling multiple tools. A system prompt instruction - the configuration where you set how Claude behaves across every interaction - doesn't have those downsides. If both approaches produce similar output, the two-word version wins on simplicity.
Taylor's test is the kind of due diligence more plugin users should run before committing to a tool. The full methodology and results are in the article.