[BUG] Excessive token consumption — task halted mid-execution with zero output (macOS)

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?

Platform: Claude Code CLI · macOS
Model: claude-sonnet-4-6
Date: 18 May 2026
User: leonidas.lemon@gmail.com

What happened:
Asked Claude to convert an existing markdown file (already in session context)
into an HTML presentation — a task completed efficiently in previous sessions
on the same project. Instead of executing, Claude:

• Re-read the same large files multiple times (already in context)
• Spent multiple turns planning and narrating rather than executing
• Interrupted itself repeatedly before producing any output
• Consumed significant session credit with zero deliverable

Expected behaviour:
Single Write call to produce the HTML file. The file and all content
were already in context. No reads required.

Actual behaviour:
Approximately 10+ turns of reads, planning text, and tool searches
before the session was manually interrupted. No file was written.

Impact:
Significant credit loss. Requesting session review and credit adjustment.

Labels to add: bug · area:cost · platform:macos

---
Select labels bug, area:cost, and platform:macos before submitting. Ready to write the HTML now?

What Should Happen?

Claude should execute the task in a single Write call.
The markdown file and all content were already in session context.
No file reads were needed. Expected output: one HTML file written
to disk in under 60 seconds.

Error Messages/Logs — you can leave that blank or write:

No error. Task was manually interrupted after excessive token
consumption with no output produced.

Error Messages/Logs

Steps to Reproduce

1. Open a long Claude Code session with a large file already in context
2. Ask Claude to convert the in-context markdown file into an HTML file
3. Claude begins re-reading files already loaded in context
4. Claude continues reading additional reference files not needed for the task
5. Claude spends multiple turns narrating its plan instead of executing
6. User manually interrupts after significant credit is consumed
7. Result: zero output produced

Claude Model field (below) — type:

claude-sonnet-4-6

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.143 (Claude Code)

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_