Why? I was enjoying reading Claude's reasoning as it works through problems, but Claude Code only lets you see the thinking at certain points and it's impossible to keep track across muliple Claude Code instances. I created a quick CLI that let me read the thinking as it happened and immediately found it to be suprisingly useful (and fun):

* I can catch it making silly assumptions and interject before it goes off the rails * It's been a great way to learn how to improve prompting - it will be quite obvious when Claude has the context it needs and when it does not * Claude will think things that it does not output and sometimes those things are worth reading * Watching Claude work through problems often gives me inspiration or ideas about architecure, dependencies, etc.

I've actually might be spending more time watching watch-claude-think than the actual Claude output! I now constantly nudge my sessions during large code changes or complex debugging issues because of what I read.

Claude helped a ton by wading through the minified Claude Code CLI and validating (or, often invalidating) my assumptions about how it stored message data. This all started as an exercise in getting to know Claude better - I wanted to try creating a project where all the code was "written" by Claude - but it ended up with a tool I keep using, maybe someone else here will find it useful too.