Hi, I have a problem with an agent and sub-agent where OpenAI overrides the sub-agent instructions:
AgentTools
1. I have Agent A and SubAgent B
2. Agent A invoked SubAgent B as a tool
3. SubAgent B has a set of instructions that I defined on the agent
4. When OpenAI decides to call SubAgent B, it writes its own set of instructions, so mine are ignored / never used
(Anthropic leaves instructions blank so everything works as expected)
How do I avoid this and make sure my initial instructions are preserved? Or at the very least the new instructions should append to the existing ones, not replace them.
2. Agent A invoked SubAgent B as a tool
3. SubAgent B has a set of instructions that I defined on the agent
4. When OpenAI decides to call SubAgent B, it writes its own set of instructions, so mine are ignored / never used
(Anthropic leaves instructions blank so everything works as expected)
How do I avoid this and make sure my initial instructions are preserved? Or at the very least the new instructions should append to the existing ones, not replace them.
