If I use the Github integration for Claude, and I assign it an issue it goes off and does the task, I can create a PR from that. But often it'll chew up a decent amount of my tokens.
Is there a tool that will create a kind of interactive visualisation of the session - and post that as a comment? That way I can see where my prompts need optimising.
Software engineer for 13 years. Been agentic coding for about three weeks.
Something I'm finding a bit frustrating is:
I can assign Claude a github issue and it'll go off of create a PR, but then I see that like half of my token allocation is used. I want to be able be able to see the cost of that session.
If I'm using Claude Code in my IDE/terminal I can watching the reasoning occur and intervene if needed, though even that gets more difficult when spawning subagents.
What I really want to be able to do is a see a graph of exactly what happened, when sub agents got spawned, what context they then provided back into their spawning agent etc.
The idea then is that I can use this to optimise my prompts.
I feel like this must fairly well trodden territory - any suggestions for tooling?
Thinking to the Github driven workflow -imagine some kind of HTML page attached to the pull request that lets you explore it.