My agent is giving standalone LLM responses, answering a lot of questions with a sinsgle "NO", and in general not knowing who it is. I'm updated to OpenClaw 2026.5.2. It feels like the openclaw system prompt is not being injected, also, the agent does not seem to rmemeber previous questions in the same session.
#Agent's have zero memory/context, even during a single conversation.
sturdy agate