Caveman-speak skill slashes Claude Code token cost

A small Claude Code skill is getting attention for a simple idea. It rewrites prompts and responses into deliberately terse language, stripping out the politeness, filler and long explanations that consume tokens. The project claims this can cut token costs by 65 percent. You can inspect the project and its approach in the GitHub repository.

The traction is notable. The repository is only about four months old, yet it has roughly 105,000 stars and has recently been gaining around 652 stars a day. That tells you developers are interested, not that every business will see the stated saving. The 65 percent figure is a claim from the project, so test it against your own work before treating it as a budgeting number.

Still, the underlying point is sound. AI costs are driven partly by how much text goes in and out of the model. If your team runs repetitive coding, research or operational tasks, verbose instructions and lengthy outputs can quietly add up. Ask whether the agent needs a friendly essay, or whether it needs a short, structured answer that completes the job.

Start with a low-risk workflow. Compare the cost, output quality and error rate between your current prompts and a terser version. Don’t optimise for cheap tokens if the result creates more human review. This is the kind of thing we build into an AI command centre, where agent runs need cost controls alongside quality checks.