Flamerix AI · 2026-10-28 · notable
Claude Code's agent status line — a test says it used 16% of spend
Flamerix AI measured Claude Code's agent_summary calls, which write a 3-5 word status for each background agent, at 16.4% of spend in its own week. A Claude Code engineer put the average at about 2% and promised a cheaper mode.
A one-line progress label for each background agent re-sends that agent's whole transcript to the model about every 30 seconds.
What is it?
Flamerix AI's write-up says Claude Code's agent_summary feature, on by default since 2.1.232, sends each working background agent's full transcript to the agent's own model, usually Opus, about every 30 seconds. The model only answers with a 3-5 word status for the agents panel. In Flamerix's team, those calls were 16.4% of spend and 19.7% of API requests in the week before they blocked them.
How does it work?
The measurement grouped Claude Code's OpenTelemetry api_request events by query_source; the summary calls do not appear in session logs on disk. With no setting to switch the feature off, the gist ships a Bun preload script (no-agent-summary.js, MIT) that replaces the summary timer with an inert one, so the panel status stops updating. The author warns it can break silently after an update.
Why does it matter?
Teams that run many background agents pay for these calls at Opus rates, and an open GitHub issue from August with four user measurements reports 18-26%. On Hacker News, a Claude Code team member (edwinarbus) said the typical share is about 2% of cost at API list prices (about 6% in the terminal), that a fix for summaries missing the prompt cache shipped in 2.1.311+, and that a cheaper mode using a smaller model is in the works.
Who is it for?
Claude Code users running background agents