I use agent orchestration flow with GitHub Copilot on my daily basis, and I was wondering if the main context window, when it comes to 50%, is it affecting the sub agents?
I mean, my rule is to always clean the context when it comes to 50%. I was working like this for the past five months, and it was working fine, so you can keep reasoning on an efficient level, and the hallucination at a minimum.
But I wonder: is it affecting the sub-agents when the main context comes to 50 or 60%, and does it affect the sub-agents context, or is it that the sub-agents always start with a clean context window?
How big is the context window for sub-agents? For example, if I use Claude Opus 4.6 as a main orchestrator and then every sub-agent is also Claude Opus 4.6.
Thanks for help.