I am using Claude design, and a lot of times it's agent is failing mid-task, with connection errors (not sure is this connection between me and agent, or between agent in Claude infra and relevant LLM model)

When these errors happen:

1. My limits have been consumed already

2. There is no resume button, there is only retry, which starts again

3. Even retry fails, then consumes my tokens.

I had some chats where I lost 80% my limits by retrying the task, with no luck of completing it.