[BUG] Automatic compression does not work, and context limit error is displayed

Environment

• Platform (select one):
• [ ] Anthropic API
• [○] AWS Bedrock
• [ ] Google Vertex AI
• [ ] Other: <!-- specify -->
• Claude CLI version:0.2.54
• Operating System: macOS 15.3.2（24D81）
• Terminal: iTerm2 Build 3.5.11

Bug Description

Today, after updating to Claude Code 0.2.54, the automatic context compression hasn't worked at all, and I'm getting the error: 'API Error: input length and 'max_tokens' exceed context limit: 187577 + 20000 > 204698, decrease input length'.

Steps to Reproduce

1. Update to Claude Code 0.2.54
2. Operate under normal usage conditions until the context becomes large
3. When automatic compression should occur, the above error is displayed

Expected Behavior

When the context size becomes large, the automatic compression feature should work, allowing operations to continue without errors.

Actual Behavior

getting the error: 'API Error: input length and 'max_tokens' exceed context limit: 187577 + 20000 > 204698, decrease input length'.

Additional Context

In previous versions, automatic compression functioned without issues.