The Shelf · Signal Library
No. 130

Claude Code's background-agent summaries burn a measured ~16% of spend (community measurement, not an Anthropic figure)

ModerateEvidence grade
Oct 29, 2026Shelved
Tools & modelsSection
sha256:a5db311d1c86…Fingerprint

a widely shared gist, backed by OpenTelemetry exports, reports that every ~30 seconds Claude Code sends each background agent's *entire transcript* to its model to produce a 3–5-word status line for the agents panel — a feature on by default since 2.1.232 with no setting to turn it off. In the author's telemetry `agent_summary` was 16.4% of all Claude Code spend (more than the main conversation at 15.7%) and 19.7% of requests, 96% of it on Opus; the gist links four independent measurements of 18–26% and ships an MIT-licensed Bun preload that suppresses the summary timer, verified across versions 2.1.257–2.1.312. On a subscription the cost is rate-limit headroom, not invoice dollars.

Read the source — GitHub GistBrowse the Shelf

Cite as — Sawdust Shelf v2026-11-05.3 · No. 130 · fingerprint sha256:a5db311d1c86…
Share image rendered from the living shelf — the number, the grade and the fingerprint travel with the card.