Auto-compact does not trigger after context exceeds 80%-100% (2.1.162)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Auto-compaction does not appear to trigger automatically in Claude Code 2.1.162.

In my session, the context usage exceeded 120%, but no automatic compaction occurred. Historically, auto-compaction would typically trigger around 75–80% context usage (or near the configured threshold), helping prevent excessive context growth.

Expected behavior:

• Auto-compaction should trigger automatically when context usage approaches the configured threshold.
• Context usage should not significantly exceed the threshold without compaction occurring.

Actual behavior:

• Context usage reached approximately 120%.
• No automatic compaction was triggered.
• The session remained functional, but context continued to grow beyond the expected limit.

It's unclear whether this is:

• A regression in the auto-compaction mechanism.
• A change in the intended behavior that is not documented.
• An issue with the context usage indicator reporting percentages above 100%.

What Should Happen?

Auto-compaction should trigger automatically when the session reaches the configured context threshold (or the default threshold if none is configured).

When the threshold is reached:

• The session should be compacted automatically without requiring manual intervention.
• Context usage should drop after compaction.
• The user should not see context usage continue growing significantly beyond the configured limit.

If context percentages above 100% are expected by design, this behavior should be clearly documented and the context indicator should accurately reflect the state of auto-compaction.

Error Messages/Logs

Steps to Reproduce

1. Start a new Claude Code session using version 2.1.162.
2. Work in the session until the context usage exceeds the expected auto-compaction threshold.
3. Continue working until the context indicator shows more than 100% usage (e.g., ~120%).
4. Observe that automatic compaction does not occur.
5. Verify that the session continues operating without reducing context size automatically.

Optional:

• Run /context and note the reported context usage.
• Run /compact manually and verify that manual compaction still works.

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.162

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

IntelliJ IDEA terminal

Additional Information

• Claude Code version: 2.1.162
• Context usage reached approximately 120% without automatic compaction.
• No manual changes were made to disable auto-compaction.
• Session remained functional despite exceeding 100% context usage.
• Manual /compact command was available and worked as expected.
• Historically, auto-compaction was observed to occur around 75–80% context usage, so the current behavior appears different from previous versions.
• It is unclear whether this is an intended change in context management, a regression in the auto-compaction mechanism, or an issue with the context usage indicator.
• Please clarify whether context usage values above 100% are expected and whether the displayed percentage is still the metric used to trigger automatic compaction.