Claude Code New Idle Compaction: after being idle for 54 minutes, automatically save Token, but context details may be lost

Article

Beating AI newsflash: Anthropic has added Claude Code Idle Compaction to Token consumption. Previously, if users left their computers for a period of time and returned to continue the conversation after the prompt cache had expired, the model had to reprocess a large amount of prior content. Now,Claude Code will automatically compress long conversations into a summary before the cache expires. When users return, only the summary needs to be reprocessed rather than the entire history, reducing the cost of subsequent calls.

This feature currently targets the terminal version. By default, it requires the session to use a 1-hour prompt cache, have context of at least 200000 Token, and keep the computer awake. Compression is triggered after approximately 54 consecutive minutes of idleness.

However, saving money comes at a cost. After compression, Claude can only continue working based on the summary and may forget specific prior instructions, technical details, and the decision-making process.October Early on, developers complained on GitHub that they were running approximately 20 long-running sessions simultaneously, and 7 of them were automatically compressed overnight, forcing them to restore critical context.

Anthropic later added an option to disable it. Users can set `"idleCompaction": false` in `settings.json`, or adjust the default 200000 Token trigger threshold.

Original link https://m.theblockbeats.info/flash/372005

Key points

Anthropic has added Idle Compaction to the terminal version of Claude Code.
By default, the feature requires the session to use a 1-hour prompt cache, have context of at least 200000 Token, and keep the computer awake.
After approximately 54 consecutive minutes of idleness, compression may be triggered.
Users can disable the feature by setting `"idleCompaction": false` in settings.json, or adjust the 200000 Token trigger threshold.
After compression, the model continues working based on the summary and may omit specific instructions, technical details, and the decision-making process.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

After a long conversation has been idle for a period of time, this feature first organizes its historical content into a summary. When the user returns, the model reads the summary rather than the full history, potentially reducing subsequent Token consumption, but the summary may not retain all details.

Why it matters to readers

It may reduce Token consumption when restoring long conversations, but omitted context details may also require users to replenish information or redo work.

If you use Claude Code for long-running tasks, be aware before leaving your computer that the session may be compressed, and check whether this feature should be disabled.

Risks and unknowns

The provided material does not include an official Anthropic announcement or complete original technical documentation.
The material does not specify the exact version to which this feature applies.
Automatic summarization may omit specific prior instructions, technical details, and the decision-making process.
The material does not provide the actual extent of the reduction in Token consumption.

Claude Code New Idle Compaction: after being idle for 54 minutes, automatically save Token, but context details may be lost

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center