Claude Code Burns a Fifth of Its Tokens on a Status Line

Starting with version 2.1.232, Claude Code — Anthropic's terminal-based coding tool — began sending the model the entire chat history of a background agent every 30 seconds. The only purpose was to generate a 3-to-5-word status line in the agent panel. No documentation mentioned this change. Independent developer Flamerix AI spotted it using OpenTelemetry, a standard tool for tracking system activity. These status-line requests accounted for 16.4% of all Claude Code spend. That is more than the main conversation itself, which came to 15.7%. The requests also made up nearly one in five API calls. Each short reply cost around nine thousand input tokens.

There is no official setting to turn this off. An open GitHub issue already contains independent measurements showing losses ranging from 18% to 26%.

A regular spending dashboard will not catch this. It tracks only the total cost, not which internal process is behind it. Anyone running several background agents can lose a fifth of their budget without noticing. If you rely on background agents, this is a reason to check detailed request logs instead of trusting the total.