[BUG] Plan Mode Regression in 2.0.51+
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?
Feedback: Plan Mode Regression in 2.0.51+
Product: Claude Code
Version: 2.0.51+ (current: 2.0.58)
Feature: Plan Mode (/plan)
---
Summary
The Plan Mode redesign in 2.0.51 removed the interactive exploration capability that was previously its core value. It now functions as a rigid implementation-plan pipeline, eliminating the space for discussion and reflection before committing to a plan.
---
Previous Behavior (< 2.0.51)
- Plan mode was an interactive space for exploration and discussion
- Users could reflect, ask questions, explore options with Claude
- No pressure to produce a deliverable
- User controlled when/if to create an implementation plan
- Supported iterative thinking: explore → discuss → refine → then plan
Current Behavior (≥ 2.0.51)
- Plan mode is now a 5-phase pipeline with forced progression:
- Launch Explore agents
- Launch Plan agents
- Review
- Write final plan to file
- Call ExitPlanMode
- System prompt enforces: "your turn should only end with either asking the user a question or calling ExitPlanMode"
- Claude is pushed to produce a written plan regardless of user intent
- No space for open-ended exploration or discussion
- User's exploratory questions are treated as implementation requests
---
Impact
Use case broken: "I want to explore and discuss options before deciding on an approach"
Example from my session:
- I asked: "Let's explore tracing and evaluation options for my project"
- Expected: Claude researches Langfuse, RAGAS, DeepEval, discusses trade-offs with me
- Actual: Claude immediately launched Plan agents and produced a full implementation plan I didn't ask for
The new design assumes entering Plan Mode = wanting an implementation plan. This removes the valuable thinking space that differentiated Plan Mode from normal mode.
---
Suggested Fix
Option A: Add an "Exploration Phase" before the planning pipeline
- Allow users to explicitly signal when they're ready to move from exploration to planning
- Don't auto-launch Plan agents until user confirms
Option B: Create separate modes
explore mode- Interactive discussion, research, no deliverable pressureplan mode- Current implementation-plan pipeline
Option C: Make the pipeline phases optional
- Let users stay in Phase 1 (exploration) indefinitely
- Only progress to planning phases when user explicitly requests
---
Environment
- Claude Code version: 2.0.58
- Model: Opus 4.5
- OS: macOS Darwin 25.1.0
---
Priority: High - This is a workflow regression that forces users out of their preferred interaction pattern.
What Should Happen?
Previous Behavior (< 2.0.51)
- Plan mode was an interactive space for exploration and discussion
- Users could reflect, ask questions, explore options with Claude
- No pressure to produce a deliverable
- User controlled when/if to create an implementation plan
- Supported iterative thinking: explore → discuss → refine → then plan
Error Messages/Logs
Steps to Reproduce
Example from my session:
- I asked: "Let's explore tracing and evaluation options for my project"
- Expected: Claude researches Langfuse, RAGAS, DeepEval, discusses trade-offs with me
- Actual: Claude immediately launched Plan agents and produced a full implementation plan I didn't ask for
The new design assumes entering Plan Mode = wanting an implementation plan. This removes the valuable thinking space that differentiated Plan Mode from normal mode.
Claude Model
Opus
Is this a regression?
Yes, this worked in a previous version
Last Working Version
2.0.50
Claude Code Version
2.0.58
Platform
Anthropic API
Operating System
macOS
Terminal/Shell
Terminal.app (macOS)
Additional Information
_No response_
This issue has 6 comments on GitHub. Read the full discussion on GitHub ↗