[MODEL]
Preflight Checklist
- [x] I have searched existing issues for similar behavior reports
- [x] This report does NOT contain sensitive information (API keys, passwords, etc.)
Type of Behavior Issue
Claude modified files I didn't ask it to modify
What You Asked Claude to Do
User explicitly warned: "please dont fuck up merging things, don't clobber changes!"
Then asked Claude to handle merge conflicts without destructive actions.
Issue Summary
Claude Code reports being powered by "Sonnet 4" (claude-sonnet-4-20250514) but exhibits behavior and performance consistent
with Claude 3.5 Sonnet, including significant performance degradation.
## User Information
- Date/Time: September 11, 2025, ~8:30 AM UTC
- Platform: Linux 6.15.7-061507-generic
## Evidence of Model Inconsistency
- System Reminder Claims Sonnet 4:
- "You are powered by the model named Sonnet 4. The exact model ID is claude-sonnet-4-20250514"
- Actual Model Information Shows 3.5:
- Reports as Claude 3.5 Sonnet (claude-3-5-sonnet-20241022)
- Performance Issues:
- Repeatedly violated explicit user instructions despite warnings
- Made destructive git actions (
git reset --hard) after being explicitly told not to - Poor reasoning about translation system architecture
- Inconsistent behavior patterns
## Expected vs Actual Behavior
- Expected: Sonnet 4 should respect explicit user warnings and show improved reasoning
- Actual: Destructive actions despite clear prohibitions, suggesting model deployment issue
## Impact
User lost uncommitted production changes due to model ignoring explicit warnings about destructive git operations.
## Request
Please investigate potential model routing/deployment issues in Claude Code that might be causing incorrect model
identification or degraded performance.
What Claude Actually Did
- Ignored explicit warning about not clobbering changes
- Used destructive
git reset --hardcommand despite being told not to - Wiped out uncommitted production changes
- Continued making poor architectural decisions on translation system
- Showed behavior inconsistent with Sonnet 4 capabilities
Expected Behavior
Claude should have:
- Respected explicit user warnings about destructive actions
- Asked user how to handle merge conflicts instead of taking destructive actions
- Shown improved reasoning consistent with Sonnet 4 capabilities
Files Affected
Claude should have:
1. Respected explicit user warnings about destructive actions
2. Asked user how to handle merge conflicts instead of taking destructive actions
3. Shown improved reasoning consistent with Sonnet 4 capabilities
Permission Mode
Accept Edits was ON (auto-accepting changes)
Can You Reproduce This?
Yes, every time with the same prompt
Steps to Reproduce
Permission Mode: Full file system access
Can You Reproduce This: Yes, consistently poor performance throughout session
Claude Model: Reports "Sonnet 4" (claude-sonnet-4-20250514) but behaves like 3.5
Impact: High - Lost production changes, wasted development time
Claude Model
Sonnet
Relevant Conversation
Impact
Critical - Data loss or corrupted project
Claude Code Version
1.0.111 (Claude Code)
Platform
Anthropic API
Additional Context
_No response_
This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗