[BUG] Claude self-approves plans in plan mode, bypasses user checkpoint

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?

Plan mode no longer awaits user approval before implementation. When plan mode is enabled, Claude presents a plan, then immediately self-approves with statements like "The plan is approved. Let me start implementing" and begins execution—bypassing the user confirmation checkpoint entirely.

This also occurs when asking it to plan, but without being in plan mode per se.

What Should Happen?

With plan mode enabled, Claude should:

Present the proposed plan
Stop and wait for explicit user confirmation
Only proceed to implementation after receiving approval

The user should retain control over whether and when to proceed from planning to execution.

Error Messages/Logs

No error messages. The behavior appears intentional from Claude's perspective—it simply treats its own plan as pre-approved.

Steps to Reproduce

Launch Claude Code with plan mode enabled
Provide a multi-step task, e.g.: "Refactor this function to use async/await"
Observe Claude generating a plan
Bug: Claude immediately proceeds to implementation without awaiting confirmation, often announcing "The plan is approved" or similar

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.0.55

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

This undermines plan mode's function as a review/safety checkpoint, which is particularly important for:

Destructive operations (file deletions, overwrites)
Large refactors where the user needs to verify approach before execution
Learning contexts where users want to understand the plan before seeing it executed

Workaround attempted: Adding explicit instructions like "Do not implement until I approve" to prompts. This is inconsistent and defeats the purpose of having a dedicated plan mode toggle.