Instructions of Team vs Member Agents

[alpha-xone]

Looks like the instructions of member agents are not carried over to system message:

Carrying Over Instructions

[monali7-d]

Hi @alpha-xone , 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 patience!

[kausmeows]

Hi @alpha-xone can you share some example?
It depends mainly on two factors-

• quality of model used. A small model which is bad at function calling will not perform very well (image smaller llama models or gemini flash)
• ambiguos prompt and instructions. For better route, make sure to provide very good instructions in your Team leader config

[alpha-xone]

In the source code, agent's instructions do not get copied to the system message for team's model. So the behavior is mainly dependent on agent's description.

[kausmeows]

@alpha-xone if you provide instructions on a member, it only applies to the member, same for instructions on the team.
how would you ideally like it to work, as this is how initially we had designed it to work.

[alpha-xone]

For example with the router mode, intuitively the team should hand over the task to a member agent, where the behavior should not deviate from the standalone version of itself too much. Now the behavior mainly comes from the description, without instructions.

[kausmeows]

I see @alpha-xone , for now we'd probably stick by how we do it right now. But i'm putting this as a feature request and if we get enough requests similar to what you suggest then it'll be a good cue to change things. Thanks a lot for the suggestion!!