Agent instruction

[yash-isr]

Does the agent's runtime behavior in Agno AI depend on the current instructions, or are the instructions fixed at initialization?

For example, if I update an agent’s instructions in memory or through code after it has started, will it adapt to the new instructions during runtime? Or do I need to reinitialize the agent for the changes to take effect?

[monali7-d]

Hey @yash-isr
thanks for reaching out and supporting Agno. I’ve shared this with the team, we’re working through all requests one by one and will get back to you soon.
If it’s urgent, please let us know. We appreciate your patienc

[Ayush0054]

Hey @yash-isr ,
Yes, you can update the instructions even after initialization using agent state.
Here’s the t doc:
https://docs.agno.com/agents/state#using-state-in-instructions

Let me know how it goes! 🙌

[yash-isr]

Hey @Ayush0054 ,
Thanks for the link — yes, I’ve used that and successfully updated the agent's instructions via the state.

The first time I call the agent after updating the instructions, it behaves as expected and follows the new instructions. But on the second call, it doesn’t respond based on those instructions anymore — instead, it seems to fall back on generic NLP reasoning (I'm using GPT-4o).

Is there a way to ensure the updated instructions persist across multiple calls?

[Ayush0054]

hey @yash-isr for persistence , have you tried using storage for this ?

[yash-isr]

Hey @Ayush0054

Let me explain my scenario:
I have a form with required fields that users need to fill out. The agent is responsible for helping users complete the form, and I'm dynamically generating the prompt using the session state.
When some required fields are missing, I expect the agent to return a clear error message indicating which fields are missing. But instead, the agent responds with a generic NLP-style message mentioning the missing fields, but not in the expected error format.
Do you have any suggestions on how to make the agent consistently return structured or clear error messages when required fields are missing?

Below is my agent prompt :

Extract only candidate details: {fields_str}

Rules:

• Return a single valid JSON object.
• If all fields are present, return:
  {{
  {json_keys},
  "actionType": "CND_CREATE"
  }}
• If any field is missing, return:
  {{
  "actionType": "ERROR",
  "error": "Missing required field(s): field1, field2, ..."
  }}
• No extra text, markdown, or explanation — JSON only.

[Ayush0054]

Hey @yash-isr,
For the expected error format, I think using structured output would be a good fit:
https://docs.agno.com/agents/structured-output

Let us know if you’re still facing the issue.
Thanks! 🙌

[yash-isr]

Hey @Ayush0054,

Let me clarify my situation:

I'm using Agno in route mode with a team (REAM) that calls other sub-teams. Within those teams, I have multiple agents, each responsible for handling different parts of the conversation and returning structured responses.

In my use case, I have a form with required fields. When users leave out some fields, I want the agent to return a clear, structured error message in JSON format like:

{
"actionType": "ERROR",
"error": "Missing required field(s): field1, field2, ..."
}

However, even though I've added the structured output rules in the prompt and tried using Agno's structured output feature, the response still comes back as a generic NLP-style message, not in the strict JSON format I expect.

The structured output doesn't persist or apply correctly when the response is returned from a sub-agent inside a team that's being routed to.

Do you have any advice or suggestions on how to make sure that structured JSON responses are returned from sub-agents within a team context?

I really appreciate any help you can provide.

[Ayush0054]

hey @yash-isr , apologies for late reply , are you still facing this issue after upgrading to latest agno version ?