Introduce a "Pause button" to the Cascade Workflow, to allow for context updates and direction fixes when the Agent is clearly not going in a direction that it should be.
Raymond Johnson
There are many times when the Thinking process will reveal what the agent is intending to do, then the output shows that it just arbitrarily made a decision about something that is critical to application flow design. If we had a Pause button, where we could highlight the reasoning of the Agent, and provide specific direction, the process could be more efficient, and less work would need to be done to repair the broad implications a single change can make in a codebase. I know we can just press stop, but that dumps the workflow the agent is doing creating a completely new execution path.
Patrick Siotti
https://windsurf.canny.io/feature-requests/p/utilidade-publica
P
Pierre-Louis Favreau
Definitely a good idea, I was also thinking, would be nice to "send context" while it's running. That way, we don't even have to pause
Today we have queued messages, but we have to wait for the end of the workflow to have the new pieces of context taken into consideration.
Raymond Johnson
Pierre-Louis Favreau, yes, ephemeral context insertion. The Agent system already has this, it just needs a user exposed interface.
Tim Billington
I just stop it and send it my feedback in a new prompt. Seems to work fine. However now the thought text is hidden, it's much harder to know when I need to stop it.
Raymond Johnson
Tim Billington, yes that is obviously a path. the problem is sometimes the Agent is going off the rails in real time and you can see it, but you need to be able to verify what it is doing to ensure you are not overreacting. A pause feature allows the opportunity to review and analyze, then choose to resume or interrupt. I agree with another commentor, having the ability to add ephemeral messaging to an active command sequence can help steer the Agent and optimize token and credit usage.
G A
WOW !! Goooood!
Jayant Mathoera
This would be a great feature!
a
andromedacgb@gmail.com
I totally agree with you... or if the agent asks me I'm in the right direction.
Guy G
This, 100 times this! Or allow us to limit how many iterations it goes through before pausing. And/or allow us to position prompts in its "path" so that it reads the prompt after the next iteration completes.