Agent

Type ID: core.agent  ·  Kind: Action  ·  Trace: AGENT

Executes an AI agent identified by its agentReferenceKey, sending it the supplied prompt. The node pauses the workflow branch and waits on the agent channel until the agent finishes and produces a response in the expected output format.

Credentials

None

Properties

Property	Key	Type	Required	Default	Possible values	Shown when
Agent	agentReferenceKey	string	Yes	—	any string (the reference key of the agent to execute)	Always
Prompt	prompt	string	Yes	—	any string (the prompt to send to the agent)	Always
Auto-Close Session	autoCloseSession	boolean	No	true	true / false — close the session automatically after the agent responds, or wait for manual user intervention	Always
Expected Output Format	outputFormat	json	No	JSON Schema with status and response string fields (both required)	a JSON Schema object defining the expected output structure	Always

Notes

• You always provide agentReferenceKey. During deploy and dev-sync the platform resolves it to the concrete agent's agentId internally.
• Parameters are resolved through the resolver engine, so expressions are supported in property values.
• execute() throws if the agent reference key or prompt resolve to empty.
• The node does not return data immediately. It returns a wait descriptor on the agent channel carrying the resolved agentReferenceKey, prompt, autoCloseSession, and parsed outputFormat, plus the input context (immediate items, immediate-by-source, and lineage-by-source). The workflow resumes when the agent produces its response.
• outputFormat may be supplied as a JSON string or an object; a string is parsed before being passed along. The default schema marks status and response as mandatory fields.