The issue is that Retell evaluates transition conditions only after the user speaks again, so prompt-level instructions telling the agent to "immediately transition" won't override that turn-taking loop on their own. Are you using a custom LLM or Retell's native LLM for this agent?