Problem: Equation-type transition conditions on conversation and subagent nodes are only evaluated after a user turn (or immediately after the agent’s turn if Skip Response is enabled — but Skip Response is an all-or-nothing node setting that also removes the ability to wait for user input on other edges from that same node). For a node that legitimately needs both behaviors — most turns wait for and process a user reply, but one specific state (e.g. a variable flipping true mid-loop) should exit immediately without needing another user turn — there’s currently no way to express that. The variable can be correctly set the moment a tool call returns, but the edge watching it doesn’t get re-checked until the caller happens to speak again, which may never happen if the caller has nothing left to say.
Concrete impact: We hit this exactly with a multi-round qualification subagent. A tool call sets {{done}} to "true" mid-call. The exit edge is a correctly-matching equation condition, but it isn’t re-evaluated until the caller’s next turn. If the caller doesn’t reply again (because our own script correctly told the agent not to prompt a further response at that point), the call sits idle until the account’s generic inactivity timeout kills it, several minutes later — even though the data needed to route the call correctly was available immediately.
Request: Add a per-edge or per-node option to evaluate equation conditions continuously (or at minimum, immediately whenever the referenced dynamic variable changes value), independent of user/agent turn boundaries — without requiring the node to give up normal turn-based behavior for its other edges. This would let a single node both (a) wait normally for user replies while still collecting information, and (b) exit immediately once a specific variable condition becomes true, regardless of whether the caller says anything else.