Field guides
Write-ups based on measurements from real Claude Code session logs, not estimates. Every number in them can be reproduced on your own machine with one command.
What's actually in your Claude Code usage logs
Where the logs live, which fields carry cost, and why summing input+output tokens undercounted real usage by 633x.
What actually drives your Claude Code bill
Opus turns cost 4.5x more than Sonnet turns while using fewer tokens. 8% of turns produced 28% of cost. 3 of 32 projects produced 77%.
$ npx github:THEMANJH/agentspend-upload --report
Free, no signup, nothing uploaded — reproduces these breakdowns against your own logs.