CLAUDE_CODE_MAX_OUTPUT_TOKENS=4096 causes infinite hang during context compaction

Description

When CLAUDE_CODE_MAX_OUTPUT_TOKENS is set to 4096, the context compaction process hangs indefinitely with no output. The session becomes completely unresponsive and must be abandoned — there is no way to recover.

Environment

• OS: Windows 11 Pro for Workstations
• Environment variable: CLAUDE_CODE_MAX_OUTPUT_TOKENS=4096

Steps to Reproduce

1. Set CLAUDE_CODE_MAX_OUTPUT_TOKENS=4096 in your environment
2. Start a Claude Code session
3. Continue working until the context window fills up and compaction is triggered
4. Observe: the compaction process starts but never completes — infinite loading with no response

Expected Behavior

• Compaction should complete successfully regardless of the max output token setting, or
• If 4096 tokens is insufficient for compaction, a clear error message should be shown instead of hanging silently

Actual Behavior

• Compaction enters an infinite loading state
• No error message, no timeout, no output
• The session is permanently stuck and must be killed
• All unsaved conversation context is lost

Additional Context

• Users set low MAX_OUTPUT_TOKENS to get more concise responses and save costs. This is a valid use case that should not break core functionality.
• The hang appears to be because the compaction summary itself requires more than 4096 output tokens to generate, but instead of failing gracefully, it silently hangs forever.
• This is particularly frustrating because it happens after a long productive session — losing all that context with no warning.