claude code is great until you lose track of what it just did. you start a session, prompt your way through a refactor, accept changes, and an hour later you're staring at a working repo with no memory of which edit broke what or why you went down the path you did. cursor has checkpoints. claude code cli doesn't. that gap bugged me enough to build one. the problem the cli is fire-and-forget by design — it edits files, you accept, the conversation rolls on. if you want to ask "what did claude actually do in the last 30 minutes?" or "let me diff against where i was four prompts ago", there's nothing in the box. you can git commit between every prompt, but mid-session i almost never do, and the commit messages would be garbage anyway.…