FULL STORY
Claude Code's Auto-Compact: From Discovery to Anthropic's Clarification
Developers discovered Claude Code auto-compacts idle conversations to save quota, but complaints of fast quota burn on long tasks prompted Anthropic to clarify how the auto-compact mechanism actually works.
2026-10-09 ~ 2026-10-11 · 2 episodes · 7 posts
Episode 1 · Claude Code now auto-compacts idle conversations to save cache credits (2026-10-09, 3 posts)
Users report that Claude Code now automatically runs /compact on idle conversations before the prompt cache expires, preventing costly cache misses involving up to 600K tokens.
- Claude Code auto-compacts while idle, before your prompt cache expires to save usage limits — JeremyNguyenPhD · 2026-10-09
- Claude Code Now Auto-Compacts Conversations Before Cache Expires — Mrk_J · 2026-10-09
- Claude Code now auto-runs /compact while idle to dodge expensive cache misses — Neither-Eye-8906 · 2026-10-11
Episode 2 · Anthropic Clarifies Claude Code Auto-Compact: Default 967K Threshold Burns Through Quota (2026-10-10, 4 posts)
Anthropic engineer Lydia Hallie clarified that Claude Code's auto-compact replaces the full conversation with a summary rather than keeping 1M tokens in background, and cache hits still count toward usage. The fast quota drain stems from the default 967K compaction threshold, so lowering it to around 400K can save quota.
- Claude engineer clarifies auto-compact: full summary replaces context, cache reads still burn quota — lydiahallie · 2026-10-10
- Claude Code auto-compact clarified: context is replaced by a summary, not kept in full — YashasGunderia · 2026-10-10
- Claude Code quota drain traced to 967K default compact threshold; set /autocompact 400k to save — dotey · 2026-10-11
- Claude Code quota drain explained: auto-compact waits until ~967K tokens by default — udmrzn · 2026-10-11