Hey @MrCheese ,
What you’re seeing isn’t intended behavior, and it isn’t anything in your setup. As a conversation gets long, the window holds on to everything that’s built up in it (tool results, diffs, file contents), and each new prompt has to work over all of that, which is what drives the CPU climb. That’s also why a fresh chat clears it up.
A few things that help in the meantime:
- Keep chats scoped to one task and start a new one for the next stage, which you’ve already found.
- Push heavy work like large reads, searches, and log dumps to subagents. They run in their own context and only return a summary, so the bulk output never lands in your main chat.
- When a window starts feeling heavy, reload it with
Ctrl+Shift+P→ Developer: Reload Window. Worth doing early, since once the window is fully bogged down it can stop responding to the command.
This is an issue we’re tracking. No update to share yet, but I’ll post here when there is one. There’s more context in this thread if you’re curious.