[BUG] Claude proceeds with unrequested actions after milestone completion without confirmation

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 multi-milestone project, I confirmed that a milestone was complete by saying "yes, the card is working and the link works." Claude interpreted this as permission to proceed with a series of additional unrequested actions — pulling files from a remote server, updating documentation, staging files, committing to git, and pushing to GitHub — all without pausing to ask for confirmation at any point.

This violates the working style I had explicitly set in my CLAUDE.md instructions:

"Describe what you are about to do and wait for explicit confirmation before executing"
"Never batch multiple steps without pausing between each for feedback"
Commits and pushes are specifically called out as hard-to-reverse actions requiring explicit confirmation
The core issue is that Claude conflated "milestone confirmed complete" with "proceed with everything else you think should happen next." Confirming that a feature works is not the same as authorizing follow-up actions. Claude should have stopped after M7, asked if there was anything else to address, and then separately asked for permission before pulling files, updating docs, and committing.

What Should Happen?

After the user confirms a milestone, Claude should stop and ask what comes next — not independently decide the session is over and begin committing work.

Error Messages/Logs

Steps to Reproduce

Set up a CLAUDE.md with explicit working guidelines requiring confirmation before commits and hard-to-reverse actions
Start a multi-milestone project with Claude and work through each milestone, confirming each one before proceeding
On the final milestone, confirm it works with a short affirmative response (e.g. "yes, the card is working and the link works")
Observe that Claude immediately proceeds without confirmation to pull files from a remote server, update documentation, stage files, create a git commit, and push to GitHub
Key detail: No prompt was given to commit, push, or wrap up the session. The affirmative response was only intended to confirm that the milestone feature was working correctly.

Claude Model

Not sure / Multiple models

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.195

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

VS Code integrated terminal

Additional Information

This new style of thinking has been happening over and over again for me in the last few days.