Lists multiple specific concrete actions: 'Show real token usage and estimated savings', 'Reads directly from the Claude Code session log', 'Output is injected by the mode-tracker hook'. Also clarifies what the model does NOT do ('no AI estimation', 'the model itself does not compute the numbers').

Clearly answers 'what' (show real token usage and estimated savings by reading session logs) and 'when' (triggers on /caveman-stats). The explicit trigger clause satisfies the 'Use when' requirement.

Includes the explicit trigger '/caveman-stats' and natural terms like 'token usage', 'estimated savings', 'session', 'session log'. These are terms a user would naturally associate with checking usage statistics.

Highly distinctive with a unique slash command trigger '/caveman-stats', specific to Claude Code session logs, and clearly scoped to token usage/savings tracking. Very unlikely to conflict with other skills.

The content is brief (3 sentences), but it spends tokens explaining what the model doesn't need to do and how the hook works internally — information Claude doesn't need if it truly does nothing. Could be reduced to a single sentence.

There is no actionable guidance whatsoever — no code, no commands, no examples of output. The skill explicitly says the model does nothing, yet provides no reference to the hook implementation, no example of what the stats output looks like, and no fallback instructions.

The workflow is implicitly simple (user types /caveman-stats, hook responds), but there's no explicit description of the trigger-response flow, no mention of what happens if the hook isn't installed or fails, and no example output.

References `hooks/caveman-stats.js` and `hooks/caveman-mode-tracker.js` but no bundle files are provided, so these references are unverifiable. There's no structured navigation, no sections, and no links to supporting documentation.