Thanks for reporting this. Two separate issues here that need different approaches.
First, please update to v1.5.5 (latest) and retest both issues.
For the Claude path confusion: Can you share a sample chat showing the misspelling? Does this always happen, or just in certain contexts? Try starting a fresh chat that mentions “zebrad” from the beginning to see if it still gets confused.
Consider adding project-specific context using Cursor. Create a .cursor/rules file explaining your project structure and the correct “zebrad” spelling. This gives Claude persistent context about your codebase.
For the Gemini authorization errors: This is Gemini’s content filtering being overly strict. Try rephrasing your requests differently, or use Claude models for those specific text replacement tasks. More info on this issue here
Please let us know if these issues persist after updating and trying these approaches.