The directive message text to surface to the AI agent
How strongly the AI should follow this directive:
low: Nice to follow, can be ignored if the AI has a better approachmedium: Should follow unless there's a clear reason not tohigh: Must follow in most casescritical: Must always follow, no exceptionsThe kind of directive:
instruction: Direct command to perform a specific action or follow a specific pathconstraint: Restriction on what the AI must NOT docontext: Important background information that should influence the AI's decisionssuggestion: Optional recommendation the AI may choose to follow or ignore
Represents a directive from an action to an AI agent. Unlike
Message(which is informational and meant for display or logging), directives are structured instructions that the agent framework surfaces as explicit guidance the AI must consider.Actions return directives when they need to steer agent behavior — for example, telling the agent which action to call next, constraining it from taking a particular path, or providing critical context that should influence its decision.
Example