Claude Code bug: idle auto-compaction at 38% context forces users to pay twice
ywy50 · ghdev · 2026-10-16
- Claude Code issue #102666 reports that idle auto-compaction fires far below the context limit — in one 1M-token session it triggered twice, at 38% (384,502 tokens) and 67% (674,652 tokens) usage, while the session was idle only because the model was waiting for a user decision.
- The summary kept numbers but dropped the reasoning behind a recommendation; thinking blocks in the .jsonl log are mostly empty (1,385 of 1,680), so nothing could be recovered.
- The user had to re-read code, re-run measurements, and re-derive the recommendation — every step billed again. The compaction call itself took 111 seconds, and recovery cost far more.
- Proposed fixes: never compact while idle below a high context share (or let users disable idle compaction), and preserve reasoning behind open questions.
More from coding & agent
- Claude Code Pushes Urgent Security Update Across All Installations — umeshonai · 2026-10-23
- Claude cut a 2.6-hour video export to 20 minutes — but only after being told to rebuild the server — byhewar · 2026-10-23
- 3 flight-booking rules my AI agent checks that I'd never do by hand — Efistoffeles · 2026-10-23
- OpenJet claims Qwen3.8-27B coding runs 1.9x faster than llama.cpp on one RTX 3090 — JeremyCMorgan · 2026-10-23
- Building an autonomous SEO agent in Hermes with three skills: programmatic pages, Keywords Everywhere, DataForSEO — EXM7777 · 2026-10-23
- Building a sports betting AI agent with onchain execution and human approval by default: architecture and lessons — Dry-Comfortable-2678 · 2026-10-23