caveman-stats

Report actual input and output tokens from the Claude Code session JSONL log.

1|Updated Apr 27, 2026

Search Tags:#Claude Code #developer tooling #caveman #token tracking #session receipts #prompt efficiency

Ready to run

npx skills add https://github.com/dwaynimay/IoTProject --skill caveman-stats-dwaynimay

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: caveman-stats
Source: https://github.com/dwaynimay/IoTProject/tree/main/.agents/skills/caveman-stats
Command: npx skills add https://github.com/dwaynimay/IoTProject --skill caveman-stats-dwaynimay

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It eliminates uncertainty about whether your prompting strategy is actually reducing token usage by showing the real input/output tokens consumed during the current session.

Core Features & Use Cases

Reads real session receipts: Extracts token counts directly from the Claude Code session JSONL log on disk rather than estimating.
Reports savings vs baseline: Presents baseline token totals alongside “caveman” token usage and saved amounts.
Instant on-demand output: Hooks into the mode-tracker and returns formatted stats when you invoke /caveman-stats, including a lifetime-savings suffix for the statusline badge.

Quick Start

Run /caveman-stats to display the current session’s input/output tokens and savings based on the on-disk session log.

Frequently Asked Questions about caveman-stats

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I track real Claude Code token usage in my current session?▼

View real Claude Code token usage by running /caveman-stats, which reads actual input and output tokens directly from the local JSONL session log on disk to display formatted stats and a lifetime-savings suffix for the statusline badge.

Can I validate prompt efficiency improvements without model-based estimation?▼

Validate prompt efficiency without model-based estimation by extracting real token counts directly from the Claude Code session JSONL log on disk, reporting baseline totals alongside actual usage and saved amounts.

Does caveman-stats work with interactive development sessions in Claude Code?▼

Caveman-stats works with interactive development sessions in Claude Code by hooking into the mode-tracker and injecting a blocked-decision response containing formatted stats when you invoke the /caveman-stats command.

What is the best way to see Claude token savings instantly during a coding session?▼

See Claude token savings instantly by running /caveman-stats, which compares baseline token totals against actual caveman token usage from the session log and presents the saved amounts on demand.

How do I display a lifetime token savings badge in the Claude Code statusline?▼

Display a lifetime token savings badge in the Claude Code statusline by triggering the /caveman-stats hook, which formats stats from the JSONL session log and persists a lifetime-savings suffix for the statusline badge.