RTK hook appears to increase LLM token consumption in Claude Code

Summary

After installing RTK with the automatic Claude Code hook, I consistently reach my Claude usage limits much faster than before.

This is unexpected because RTK’s goal is to reduce token consumption.

Environment

• macOS
• Claude Code
• RTK installed using the automatic global hook (rtk init -g)

Expected behavior

RTK should reduce the overall number of tokens consumed during a development session.

Actual behavior

After enabling the automatic hook:

• Claude reaches my usage limits significantly faster.
• Development sessions appear to consume more tokens overall.
• Claude seems to perform more tool calls than before.

My hypothesis

I suspect that the summarized output generated by RTK sometimes removes information that Claude would normally use for reasoning.

As a result, Claude appears to compensate by:

• executing additional shell commands,
• reopening files,
• performing more searches,
• or requesting more context,

which may ultimately consume more tokens than the original unfiltered output.

In other words, while each individual command output may be smaller, the total amount of reasoning and tool usage across the session appears to increase.

Request

Would it be possible to:

• provide telemetry comparing total session token usage (not only per-command reduction),
• add a debug mode showing when RTK removed information that may trigger additional tool calls,
• or provide recommendations about which commands should not be filtered in Claude Code?

I’d be happy to help test or provide additional information if needed.

Thank you

Start by reproducing the behavior with the automatic global hook configured by rtk init -g, comparing total session usage and tool-call counts with and without RTK. The issue mentions no source files or tests, so first identify the hook and command-filtering entry points. Done would require narrowing the cause and defining a concrete, validated change or recommendation.