// LOCAL · NOTHING UPLOADS · NO INSTALL NEEDED

How many tool calls did Claude Code run before your next message?

Mine ran 274 before I said anything. This counts yours — from your own log, in this browser tab.

Count mine in the browser drop a .jsonl session log here, or click to pick one

Usually in ~/.claude/projects/. Nothing uploads, nothing installs — the file is read in this tab.

Prefer the terminal?

A REAL RESULT 274 tool calls

longest run before the next human message one session counted locally

Longer isn't better: a long run may be garbage you haven't stopped yet. The number describes the collaboration, not the model's quality. How it is counted →

What this number is, and is not

Counting a log in your browser reports observed activity only: tool calls, runs between your messages, and how those runs ended. It runs no command and checks no outcome, so it never claims Claude caused anything. If you want a result checked as well, the CLI can run your own test command and stamp the receipt with its current pass or fail — that path is opt-in and lives on the activity page.

Check a result tooRun your own test command with the CLI and stamp the receipt with its outcome. Read the methodSee the limits and the protocol for repeated comparisons. Why we built thisExplore 30 public agent verdicts and their evidence.