The beta feature for iterating on lint errors should work on every response from the AI, not just the first few!
For TypeScript specifically, you might be able to tweak the solutions the agent provides with some rules instructing it on how to behave.
Let me know if you’re still seeing weird behavior with specific examples