Chat history compression increases the token usage

[ShammiAnand]

Chat history compression increases the token usage often even when the history is over 80% full

The following example from the bug is for a small history but the issue is not limited to that case.

What happened?

on doing /compress

Chat history compressed from 1957 to 2247 tokens.

What did you expect to happen?

on doing /compress

Chat history compressed from 1957 to 2247 tokens.

Client information

• CLI Version: 0.1.9
• Git Commit: 34935d6
• Operating System: linux v22.8.0
• Sandbox Environment: no sandbox
• Model Version: gemini-2.5-pro
• Memory Usage: 175.3 MB

Login information

No response

Anything else we need to know?

No response

[jacob314]

This is somewhat expected behavior if the history is very small. We should raise this to a P1 if you are able to reproduce this when the history is large.
For the case of very small, I'd be happy to accept a pull request that makes this case less confusing for people.
Would be reasonable to say something like "Skipping compression for the small history as the process would have increased its size."

[ShammiAnand]

@jacob314

This is somewhat expected behavior if the history is very small. We should raise this to a P1 if you are able to reproduce this when the history is large.
For the case of very small, I'd be happy to accept a pull request that makes this case less confusing for people.
Would be reasonable to say something like "Skipping compression for the small history as the process would have increased its size."

Makes sense. I have made the suggested changes here: #4404