[BUG] User billed in tokens for the assistant's own mistakes — refund request and product fix

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?

Assistant errors in this session (3):

Unnecessary file reads — the assistant read 4 files (~870 lines) when only 1 (~230 lines) was relevant to the task. ~640 lines of reading wasted (estimated ~8–11k tokens).
Left code in a broken state — during an interrupted edit, it removed an import that was still in use, leaving a source file failing to compile; it had to be reverted (more tokens spent on the fix).
Unrequested actions — it implemented code changes when I had only asked a yes/no confirmation question. It edited files with no instruction from me to do so.
My core question:
Why do I pay, in tokens, for the assistant's mistakes? When the assistant reads what it shouldn't, breaks the code, or acts without my request, the cost of those tokens comes out of my pocket. There is no distinction between useful work and an assistant error, and no automatic refund.

What I'm requesting:

Refund/credit for the tokens consumed by the errors above.
Product fix: tokens consumed by clear assistant errors (unrequested actions, broken code, redundant reads) should not be billed to the user.
Environment: Claude Code (VSCode extension), model Opus 4.8 (1M).

What Should Happen?

Assistant errors in this session (3):

Unnecessary file reads — the assistant read 4 files (~870 lines) when only 1 (~230 lines) was relevant to the task. ~640 lines of reading wasted (estimated ~8–11k tokens).
Left code in a broken state — during an interrupted edit, it removed an import that was still in use, leaving a source file failing to compile; it had to be reverted (more tokens spent on the fix).
Unrequested actions — it implemented code changes when I had only asked a yes/no confirmation question. It edited files with no instruction from me to do so.
My core question:
Why do I pay, in tokens, for the assistant's mistakes? When the assistant reads what it shouldn't, breaks the code, or acts without my request, the cost of those tokens comes out of my pocket. There is no distinction between useful work and an assistant error, and no automatic refund.

What I'm requesting:

Refund/credit for the tokens consumed by the errors above.
Product fix: tokens consumed by clear assistant errors (unrequested actions, broken code, redundant reads) should not be billed to the user.
Environment: Claude Code (VSCode extension), model Opus 4.8 (1M).

Error Messages/Logs

Steps to Reproduce

Assistant errors in this session (3):

Unnecessary file reads — the assistant read 4 files (~870 lines) when only 1 (~230 lines) was relevant to the task. ~640 lines of reading wasted (estimated ~8–11k tokens).
Left code in a broken state — during an interrupted edit, it removed an import that was still in use, leaving a source file failing to compile; it had to be reverted (more tokens spent on the fix).
Unrequested actions — it implemented code changes when I had only asked a yes/no confirmation question. It edited files with no instruction from me to do so.
My core question:
Why do I pay, in tokens, for the assistant's mistakes? When the assistant reads what it shouldn't, breaks the code, or acts without my request, the cost of those tokens comes out of my pocket. There is no distinction between useful work and an assistant error, and no automatic refund.

What I'm requesting:

Refund/credit for the tokens consumed by the errors above.
Product fix: tokens consumed by clear assistant errors (unrequested actions, broken code, redundant reads) should not be billed to the user.
Environment: Claude Code (VSCode extension), model Opus 4.8 (1M).

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

claude-opus-4-8[1m].

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_