Cut coding-agent response tokens with Caveman mode
Install Caveman as a Claude Code style skill when coding agents need terse, technically accurate responses that reduce output-token waste.
npx skills add agentskillexchange/skills --skill cut-coding-agent-response-tokens-with-caveman-mode
Use Caveman when an operator wants a coding agent to keep technical accuracy while compressing its conversational output. The agent can configure the Caveman skill, invoke it with the documented trigger, select the compression level, and review the resulting replies so explanations, status updates, and fix notes stay brief without changing code, commands, or error text. Invoke this instead of relying on normal agent verbosity when the bottleneck is repeated token-heavy status and explanation output in Claude Code-style coding sessions. The boundary is response-style compression for coding agents during an active workflow, not a general token optimizer, model router, coding agent replacement, benchmark, or prompt-engineering library.
What this skill actually does
Inputs and prerequisites: Claude Code or compatible coding-agent skill/plugin environment.
Setup notes: Follow the repository INSTALL.md for the target agent, then invoke Caveman with the documented /caveman trigger or equivalent natural-language trigger. Review any installer before running it in a managed environment.
Source and verification boundary: use https://github.com/JuliusBrussee/caveman as the canonical reference before running the workflow; keep commands, API calls, CLI usage, and generated outputs reviewable against that upstream source.
Framework fit: publish this as a Claude Code workflow only when the operator can invoke the documented toolchain directly, rather than treating the upstream project as a generic product listing.