[BUG] Claude Code wasted tokens reading 15+ irrelevant files before making any change

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?

During a session where I asked Claude Code to fix a layout issue, it read more than 15 files unrelated to the task before making any actual change. These included backend files (notifications.py, issue_service.py, issues.py), routing files, Docker configuration, system processes (ps aux), and multiple hook files — none of which were related to the layout problem reported.

The agent spent the entire session on analysis and investigation instead of identifying and fixing the issue directly. By the time any change was made, the session had consumed a significant number of tokens with no meaningful output.

Expected behavior: When asked to fix a layout, the agent should identify the relevant files quickly (2–3 reads maximum) and apply the fix.

Actual behavior: The agent read 15+ irrelevant files, ran multiple shell commands unrelated to the task, and only made a 2-line change (min-h-0) at the very end — after the user expressed frustration multiple times.

Impact: Wasted tokens, wasted time, problem still unresolved.

What Should Happen?

During a session where I asked Claude Code to fix a layout issue, it read more than 15 files unrelated to the task before making any actual change. These included backend files (notifications.py, issue_service.py, issues.py), routing files, Docker configuration, system processes (ps aux), and multiple hook files — none of which were related to the layout problem reported.

The agent spent the entire session on analysis and investigation instead of identifying and fixing the issue directly. By the time any change was made, the session had consumed a significant number of tokens with no meaningful output.

Expected behavior: When asked to fix a layout, the agent should identify the relevant files quickly (2–3 reads maximum) and apply the fix.

Actual behavior: The agent read 15+ irrelevant files, ran multiple shell commands unrelated to the task, and only made a 2-line change (min-h-0) at the very end — after the user expressed frustration multiple times.

Impact: Wasted tokens, wasted time, problem still unresolved.

Error Messages/Logs

Steps to Reproduce

During a session where I asked Claude Code to fix a layout issue, it read more than 15 files unrelated to the task before making any actual change. These included backend files (notifications.py, issue_service.py, issues.py), routing files, Docker configuration, system processes (ps aux), and multiple hook files — none of which were related to the layout problem reported.

The agent spent the entire session on analysis and investigation instead of identifying and fixing the issue directly. By the time any change was made, the session had consumed a significant number of tokens with no meaningful output.

Expected behavior: When asked to fix a layout, the agent should identify the relevant files quickly (2–3 reads maximum) and apply the fix.

Actual behavior: The agent read 15+ irrelevant files, ran multiple shell commands unrelated to the task, and only made a 2-line change (min-h-0) at the very end — after the user expressed frustration multiple times.

Impact: Wasted tokens, wasted time, problem still unresolved.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

claude-sonnet-4-6

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_