Claude Code regularly sends each background agent’s transcript to its model
Why CEREBRO kept it
Background agent transcripts sent to model; token and context cost.
The text below is an automated extraction of the article at https://gist.github.com/flamerix-ai/c3905705cc4205e5de3011fa2e74efce, stored verbatim in the public cerebro-vault repository. Copyright remains with the original publisher (gist.github.com).
by Flamerix AI · MIT License Tested on Linux, Claude Code 2.1.257–2.1.312 (see Tested versions). The binary is not modified (BUN_OPTIONS=--preload). An update can silently undo it, so see Verify. curl -fsSL https://gist.githubusercontent.com/flamerix-ai/c3905705cc4205e5de3011fa2e74efce/raw/no-agent-summary.js -o ~/.local/share/no-agent-summary.js claude daemon stop --any BUN_OPTIONS="${BUN_OPTIONS:+$BUN_OPTIONS }--preload $HOME/.local/share/no-agent-summary.js" claude To make it permanent for every Claude Code process, see Install. Every ~30 seconds, for every background agent, Claude Code sen
Community take
One team reports agent_summary calls every ~30s per background agent used 16.4% of weekly token spend, but a commenter says it's the agents-panel summary rather than the status line, and prompt caching may make the repeated transcript reads cheap.
Backlinks
Appeared in 1 briefing
Related
Shares tags: ai/llm-mechanics · cerebro/signal
Also from gist.github.com
Only signal from gist.github.com so far.