claude-code - 💡(How to fix) Fix Claude Code VS Code : ~181k token system prompt on new chats causes immediate "Prompt is too long"

Preflight Checklist

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

What's Wrong?

In the Claude Code VS Code extension, every new chat with model claude-sonnet-4-6 starts with an extremely large system prompt (~181.7k tokens), leaving almost no space in the 200k context window.

As a result, even a single short user message immediately triggers "prompt is too long" / context overflow, or shows ~98% context usage before I have any real conversation.

This happens even after clearing all chats and deleting memory.

Observed behavior

Context usage for a brand-new chat (example values from my screenshot):

• Model: claude-sonnet-4-6
• Total: 195.7k / 200.0k tokens (98%)
• System prompt: 181.7k tokens (~90.9%)
• System tools: 11.5k tokens (~5.7%)
• Skills: 2.3k tokens (~1.2%)
• Messages: 119 tokens (<0.1%)
• Autocompact buffer: 33.0k tokens (~16.5%)

Because the system prompt alone is ~181k tokens, there is essentially no free context for normal usage. I repeatedly get "prompt is too long" or hit the context limit almost instantly in any new chat.

This occurs:

• With all previous chats deleted.
• With memory cleared.
• In an empty folder with no project files opened.

Environment

• OS: Microsoft Windows 11 Pro 10.0.26200
• VS Code version: 1.123.0
• Claude Code extension version: Claude Code extension version: [email protected]
• Model: claude-sonnet-4-6 and claude-haiku-4-5
• Other extensions that might interact with Claude: "none"

Additional context

• I have cleared all chats, deleted memory, deleted extensions and plugins, and restarted VS Code.

• I see roughly the same ~181k system prompt size on every new chat, even in an empty workspace and with only a single short message.

• Screenshot of the context usage panel is attached below.

What Should Happen?

• A new chat should start with a much smaller system prompt (typical sizes are on the order of tens of thousands of tokens, not ~180k), leaving the majority of the 200k context window available for user messages, tools, and compacted history.
• Context usage in a fresh chat with a single short message should be well below 50%, and should not immediately trigger "prompt is too long".

Error Messages/Logs

Steps to Reproduce

1. Open VS Code.
2. Open a fresh workspace (can be an empty folder with no files, with all history deleted).
3. Open the Claude panel and start a new chat.
4. Select model: claude-sonnet-4-6 or claude-haiku-4-5
5. Type a very short message (e.g. "test").
6. Open the context usage panel.

Claude Model

Sonnet (default)

Is this a regression?

I don't know

Last Working Version

No response

Claude Code Version

2.1.163 (Claude Code)

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

VS Code integrated terminal

Additional Information

Co-written by Claude.