Caveman
Give coding agents shorter replies with Caveman, an installable skill with optional local tooling for compressing the information an agent reads.
What is Caveman?
Caveman adds concise response instructions to coding assistants. Its skill modes shorten explanations while keeping code, commands, paths, numbers, and negations intact. You can enable the standard mode or choose stronger compression, then turn it off when you need fuller explanations. The repository also includes a separate local proxy and command-line tools that compress noisy inputs such as logs and test output. Installing the skill alone changes output style; it does not enable the proxy. Start with the skill in a supported assistant, or follow the proxy setup guide if input compression is your goal. Token savings depend on the model, task, caching, and billing plan. Treat the published benchmarks as particular experiments, rather than a guarantee for your sessions. The CLI has usage telemetry enabled by default and documents an opt-out.
Best for
Developers using coding assistants who prefer concise explanations and want to compare optional input compression on their own work.
Not for
People who need extensive teaching or expect shorter replies to reduce a flat subscription or per-request bill.