Bug: `Agent` `instructions` field is not required and accepts non-string types

[MuhammadHamidRaza]

Problem Description

The Agent class in the SDK currently exhibits two critical issues related to its instructions field:

1. Optional instructions Field: The instructions parameter for the Agent class is currently optional. This design choice allows for the creation of agents without any defined purpose or behavior, leading to unpredictable outcomes and making the agent's functionality ambiguous. An agent's instructions are fundamental to its operation and should be mandatory for proper instantiation.

2. Acceptance of Non-String Types for instructions: The instructions field currently accepts data types other than string (e.g., integers, booleans, lists, dictionaries). This violates type safety principles and the inherent purpose of instructions, which are intended to be natural language directives for the underlying Language Model (LLM). Passing non-string values can lead to runtime errors, misinterpretations by the LLM, or silent failures that are difficult to debug.

Expected Behavior

1. instructions should be a required parameter: The Agent constructor should enforce that instructions is always provided upon instantiation.
2. instructions should strictly enforce str type: The instructions parameter should only accept a string data type. Any attempt to pass a non-string value should result in a TypeError or a similar clear validation error at the point of instantiation.

Proposed Solution

To resolve these issues and make the Agents SDK more robust and intuitive for all users, I propose the following:

1. Make instructions a Mandatory Setting: The Agent should always require instructions when it's created. This ensures every agent has a clear purpose from the start.
2. Ensure instructions Only Accepts Text: Implement a check to make sure that the instructions are always provided as a string (text). If anything else is given (like numbers or other data types), the SDK should immediately show an error, guiding the user to provide valid input.

Addressing these points would significantly enhance the robustness, predictability, and overall user-friendliness of the Agents SDK, especially for developers integrating the library.

[seratch]

Indeed, having instructions is highly recommended but it's okay to run an agent with it when you want to try something quickly. Also, the validation will be added by #998 so please check the progress there.

[MuhammadHamidRaza]

It’s absolutely fine if the user doesn’t provide instructions — no issue at all. But when they do, it should either be a string or a callable; nothing else should be allowed. Right now, other types like int can still get through. If you're okay with it, I’d be happy to add this validation. Since this is an open-source project and I’m contributing elsewhere too, I just wanted to ask first before making the change.

[MuhammadHamidRaza]

Indeed, having instructions is highly recommended but it's okay to run an agent with it when you want to try something quickly. Also, the validation will be added by #998 so please check the progress there.

Please tell me can i ffix it

[seratch]

Thanks for your interest, but the linked PR is already in the final phase before getting merged. So, contributing to other ones would be appreciated.

[MuhammadHamidRaza]

Thanks for your interest, but the linked PR is already in the final phase before getting merged. So, contributing to other ones would be appreciated.

Ok thanks your reply