Add the ability to provide real-time guidance or context to an active Agent without terminating its current workflow. This would allow users to inject helpful information that the Agent can incorporate into its ongoing execution.
Use Case: When an Agent is executing a multi-step process, users often observe issues or have relevant context that could help the Agent succeed, but currently must choose between:
-
Letting the Agent continue and potentially fail
-
Stopping the entire workflow and starting over
Proposed Behavior:
-
Allow users to send messages to a running Agent
-
Agent queues the message and processes it after completing its current atomic action
-
Agent incorporates the new context and adjusts its approach accordingly
-
Workflow continues seamlessly without full interruption
-
Example Scenario: Agent is setting up a port-forward but the command fails due to an existing background process. Instead of stopping the entire Agent, the user could interject: “There’s already a port-forward running on that port - you’ll need to kill the existing process first.” The Agent would then adapt its strategy to handle the port conflict before proceeding.