[BUG] Prompting for complete change sets is honored not in advance

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?

When Claude Code proposes a source code change which involves more than one code chunk, it often just starts with an "Edit" task of just the first chunk and then subsequently proposes the remaining chunks in separate "Edit" tasks. This iterative approach sometimes is ok. But often one cannot decide whether the proposed changes as a whole are good or bad just based on the first chunk.

One can then answer "no" on the "Edit" task and use a custom slash command (or an ad-hoc prompt) with an instruction like Please re-assess and propose the entire necessary change set, all at once. which then always leads to an subsequent "Edit" task with the first and all remaining chunks together. That's great. Unfortunately, there is no possibility, neither with CLAUDE.md files nor the system prompt to give this instruction already up-front. One always has to give this instruction for a complete change set all "Edit" tasks and even within a session every time.

What Should Happen?

I would like to see that Claude Code respects such instruction for always performing complete change sets as a whole also when given in project CLAUDE.md files or globally via the system prompt.

Error Messages/Logs

Steps to Reproduce

Give Claude Code an instruction to perform any change which involves changing more than one chunk of a source file. Then you can see that it starts its first "Edit" task with just the first code chunk. Then instruct it with a prompt like "Please re-assess and propose the entire necessary change set, all at once." and see that now the "Edit" task shows the complete change set. Now try from scratch and give the instruction upfront, before any code change and observe that once Claude Code start with the first "Edit", it does not honor the instruction. You always have to give this instruction once again on the "Edit" command by answering "no" and entering the prompt.

Claude Model

Opus

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

1.0.108

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_