[BUG] Excessive token consumption — ~2000 tokens per simple query with prompt cache enabled

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?

Every simple query consumes approximately 1% of the
daily token limit (200k tokens / 5h on Pro plan),
which equals ~2000 tokens per message. This happens
with both Sonnet 4.6 and Haiku 4.5 models.

Prompt caching appears to not be working. The
CLAUDE.md file (~15k tokens) seems to be reloaded on
every request instead of being cached, causing
excessive token consumption.

settings.json has no cache-disabling configuration.

What Should Happen?

Each query should consume significantly fewer tokens
after the first message in a session, as the system
prompt and CLAUDE.md should be cached. Expected
consumption: ~200-500 tokens per simple query, not
~2000.

Error Messages/Logs

Steps to Reproduce

1. Open Claude Code CLI with a CLAUDE.md file (~15k

tokens of system instructions)

1. Ask any simple question (e.g. "what is the

difference between models?")

1. Check token usage on claude.ai account dashboard
2. Ask another simple question
3. Check token usage again

Result: each query consumes ~2000 tokens (~1% of the
200k/5h Pro plan limit)
Expected: subsequent queries should use much less due
to prompt caching

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.110 (Claude Code)

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Warp

Additional Information

_No response_