Based on your detailed bug report, this appears to be a known issue with node transitions in Conversation Flow agents, particularly after agent transfers. The documentation and community discussions provide several relevant insights and troubleshooting approaches.
Known Node Transition Issues
Your issue matches patterns documented in the community where nodes transition incorrectly or fail to evaluate conditions properly(1). Specifically, cases where “a flow like: A → B → C → D” jumps unexpectedly, or where transition conditions are met but the agent doesn’t advance(2).
Agent Transfer Behavior
When using Agent Transfer, the destination agent receives the full conversation history(3). However, the documentation notes that certain settings from the first agent persist throughout the call, while others (like language, voice, and voiceModel) reflect the currently active agent(3).
Debugging Node Transitions
For transition issues, the documentation recommends(4):
Review transition conditions - Ensure they precisely match your intended triggers. Consider prompt engineering or breaking down complex conditions into multiple simpler ones(4).
Add transition finetune examples - Provide examples to help the model understand your expectations(4)(2). For your specific case, you would provide a transcript context showing the user responding “yes” or “no” to the callback question, and specify “OK. Got It!” as the transition result(2).
Make conditions more specific - Instead of generic conditions like “User replies,” try more explicit conditions such as “User responds to question of their age” or “User indicates they want to book a meeting”(5).
Split nodes - If a node contains multiple responsibilities, split it into separate nodes to improve consistency(4).
Testing Differences
You can test agent transfer in both web call and playground(3). However, the discrepancy you’re seeing between manual testing and voice calls suggests a platform-specific issue with how the manual test interface handles post-transfer transitions.
Recommended Actions
Since your voice calls work correctly with identical configurations, this confirms your agent setup is valid. For the manual testing issue:
- Document the working voice call behavior as your baseline
- Add transition finetune examples to the “Offer Callback” node showing the exact user responses and expected “OK. Got It!” transition(2)
- Tighten the transition condition - replace “User replies” with something more specific like “User responds affirmatively or negatively to callback question”(1)
- Check for global nodes that might be interfering with transitions(1)
- Review call logs for the specific condition that fired during the incorrect transition(1)
Reporting the Bug
Since this is a platform UI discrepancy between manual testing and production calls, you should create a ticket in the bug report channel with your call ID (call_723ff620dbce28657006c3356de) and agent IDs to help the support team investigate the manual test interface behavior(6).
The contradictory debug information (showing “OK. Got It!” internally while routing to “begin”) is particularly valuable evidence for the support team to diagnose the root cause.