input_tokens across all calls. Note: with prompt caching, this only counts non-cached tokens. The full context size includes cached tokens.output_tokens (model responses) across all calls.tool_tokens / total_input — fraction of input consumed by tool results. >30% yellow, >50% red.duplicate_tool_tokens / total_tool_tokens — tool results that appeared in a previous call (wasted context). >10% yellow, >20% red.tools[] array sent every request). These are fixed overhead per call.| Tool | Tokens |
|---|
| File | Tokens |
|---|