What problem does it solve? Token-saving prompt styles are hard to evaluate because the model cannot measure its own usage. This Skill reads the current Claude Code session log directly and reports real input/output token counts plus estimated savings versus a non-caveman baseline, so you can see whether the mode actually pays for itself. ## Core Features & Use Cases - Real token receipts: Parses the on-disk JSONL session log for actual input and output token counts instead of model estimates. - Honest net accounting: Shows estimated rule overhead (default 1,250 input tokens per turn, configurable via CAVEMAN_RULE_OVERHEAD_TOKENS) and a net figure that plainly flags net-negative sessions. - Statusline badge support: Writes a lifetime-savings suffix file consumed by the statusline badge for a glanceable gross-savings summary. - Use Case: After a long coding session, run the stats command to learn that terse replies made the mode net-negative, and decide to disable it for that workload. ## Quick Start Type /caveman-stats in the current session to see real token usage, estimated savings, rule overhead, and net cost.
