Flamerix gist: Claude Code status line ate 16.4% of weekly spend

A GitHub gist by Flamerix AI measured Claude Code's agent_summary feature consuming 16.4% of one week's spend, more than the 15.7% that went to the main conversation itself. The feature's job is to generate a 3-5 word status line.

Every ~30 seconds, for every background agent, Claude Code sends that agent's entire transcript to the agent's own model, usually Opus, and asks it to "describe your most recent action in 3-5 words". Flamerix writes: "The answer only updates the agent's status line in the agents panel; it never reaches the model or your conversation." In the same run, 19.7% of all API requests were summary calls, at roughly 9,000 input tokens per output token.

The feature has been on by default since version 2.1.232, which the gist dates to August 2026, with no documented setting to disable it. The author's workaround is a ~80-line Bun preload called no-agent-summary.js that wraps setTimeout and setInterval, matches the summarizer's callback by source text, and returns an inert handle instead of arming the timer. The gist says it has been verified across versions 2.1.257–2.1.312 on Linux.

The post landed on Hacker News and drew a reply from edwinarbus of Anthropic's Claude Code team. His account differs: across users sending telemetry, agent_summary represents roughly 2% of API-list-price cost, 6% in the terminal specifically; the 16.4% figure, he said, reflects "a team running mostly background agents" and "doesn't seem like a typical session". He pointed to a cache-miss fix already shipped in version 2.1.311+ and said the team is working on "a much cheaper mode that sends short notes to a smaller model".

The thread fits a widening run of coverage about Opus-tier calls stacking up quietly inside agent loops on our Anthropic tracker. Flamerix publishes the preload under MIT and warns the match depends on current implementation details.