[BUG] Pro plan token limits exhausted after 3-4 queries (VS Code Extension v2.1.84)

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?

After updating to VS Code extension version 2.1.84 and using Claude Sonnet 4.6, the Pro plan token usage has become unsustainable. Previously, I could work for extended periods (hours) on a codebase. Currently, after sending as few as 3 to 4 queries (even simple ones like reading a file or making a small edit), the /stats command shows I have hit my usage limit (100% utilization).

The issue occurs regardless of the complexity of the query. Even basic "read this file" or "explain this function" requests seem to be consuming tokens equivalent to complex refactoring tasks. This makes the tool unusable for standard development workflows on the Pro plan.

What Should Happen?

The Pro plan should allow for a reasonable number of interactions (typically 30-50 messages per 5-hour rolling window, depending on context size). A handful of simple queries should not deplete the entire 5-hour token allocation.

Error Messages/Logs

// Running /stats after 4 queries:
Usage limit: 100% utilized.
Reset time: [approx 5 hours from first query]

Steps to Reproduce

Environment: Open VS Code with Claude Code extension v2.1.84 active.

Start Session: Run /clear to start a fresh session.

Query 1: Ask a simple, low-context question.

Prompt: @README.md What is the main purpose of this project?

Query 2: Ask for a simple edit.

Prompt: Add a comment to the top of src/index.js explaining the import statements.

Query 3: Ask for a basic status check.

Prompt: Rungit statusand tell me which files are modified.

Check Usage: Run the /stats or /cost command.

Result: Usage is reported as 80-100% utilized despite the small number of simple interactions.

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

2.1.83

Claude Code Version

2.1.84

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

VS Code integrated terminal

Additional Information

Research Context & Potential Causes:

Based on recent community reports and documentation, this issue may be linked to several factors, particularly when combined with Sonnet 4.6:

• Aggressive Context Usage: Recent updates may have changed how the agent loads context. Even if the user doesn't explicitly reference large files, the agent might be loading extensive architectural context or performing internal "research" that consumes tokens without visible output.

• Agent Teams Overhead: The Pro plan includes access to "Agent Teams" (parallel sub-agents). If the extension is inadvertently spawning multiple sub-agents for simple tasks, each with its own context window, this would multiply token consumption rapidly.

• Sonnet 4.6 Specifics: It's possible that the token accounting or context management for Sonnet 4.6 differs from older models, and the extension may not be handling it efficiently.

• MCP Server Definitions: While I have MCP servers configured (e.g., filesystem), they were present in the previous working version as well. However, recent updates to tool handling might be causing the schema definitions to count against the context window more aggressively than before.

• VS Code Integration Regression: There is growing discussion on Hacker News and GitHub regarding the v2.1.84 update causing the "context limit to fill up much faster on the same project".

Verification Commands:

To help debug, I have verified that:

• claude --version returns 2.1.84.
• Running /cost shows high token counts even for trivial interactions.
• Running /context shows that the context window is filled with system prompts and tool definitions, but does not explain why a simple query consumes the equivalent of 20+ messages in token cost.

Request for Investigation:
Please investigate the token accounting logic in the VS Code extension (v2.1.84) specifically. It appears the extension may be either:

• Sending large hidden context with every request (increasing input tokens).
• Inefficiently managing the "Agent Teams" feature on the Pro tier, causing over-consumption.
• Misreporting usage relative to the actual 5-hour rolling window limits.