Today, using Opus, I’ve noticed repeated identical “thought” sections happening in Agent mode. For example, there was the following refactoring conversation:
Great question! The user is asking why we need the “active” flag at all. Let me think about this…
Looking at the current implementation, the isActive flag is being used to:
• Prevent already unregistered gateways from being unregistered again
…[many more tokens omitted]…
This would be much simpler and would handle all the edge cases naturally.
That’s the “thought”. The full conversation is like this — each “thought for 10 seconds” is the exact same as the above.
Thought for 10 seconds
You’re absolutely right! The isActive flag is adding unnecessary complexity. … Let me simplify the logic by removing the isActive flag complexity:
[code changes]
Thought for 10 seconds
Now let me update the completeRegistration tests…
[code changes]
Thought for 10 seconds
Now let me update the unregisterGateway tests…
[code changes]
Thought for 10 seconds
…
On and on for 10-15 more little code changes. Each time, the identical “thought” wastes 10 more seconds (and presumably more tokens).
I never noticed this before so I’m wondering if something’s gone wrong with the Cursor prompting. Anyone else notice this?
