
Not directly — the output from the Agent instructions field can’t be stored in a variable. Because it’s generated by the autonomous orchestration layer, it isn’t exposed as a system or output variable. There’s no public roadmap commitment to change that.
A practical workaround
Instead, move the logic out of Agent instructions and into a Prompt action (or Generative Answers node) inside a topic, then map the result to a variable:
Topic: HandleUserQuery
├─ Trigger: On Unknown Intent (or specific phrases)
├─ Prompt action → input: Activity.Text
│ system prompt: <your former agent instructions>
│ output: Topic.AgentResponse
├─ Message node → "{Topic.AgentResponse}"
└─ Power Automate → pass Topic.AgentResponse
With that in place, Topic.AgentResponse can be used for Power Automate, logging, and conditional logic.
Other options, though less ideal
The Agent instructions field is intentionally opaque to topic flow — anything you need to capture has to live inside a topic node.