ritual-meta-human-in-loop

Mid-session workflow decisions occur when an agent encounters an architectural fork with multiple viable approaches and must prompt the user for direction. This skill enforces fork detection and budget-aware prompting to pull the user only when the choice cannot be resolved automatically.

Human-in-the-loop elicitation should be used only when an architectural fork is expensive to switch and the next step presents multiple viable approaches. It restricts interruptions to high-impact decision points, preserving workflow momentum and avoiding unnecessary derailment.

To prompt users without disrupting the session, apply canonical fork points and decision rules that evaluate fork potential, budget, and prior elicitation. This ensures minimal disruption while preserving decision quality and workflow momentum.

Automated elicitation is insufficient for architectural forks that are expensive to switch. When the next step presents multiple viable approaches that cannot be resolved by the agent alone, you must escalate safely and pull the user for a mid-session decision.

Automated decision-making fails when encountering high-impact architectural forks where the choice cannot be discovered automatically. The limitation is resolved by enforcing safe escalation and pulling the user mid-session to manage forks that are expensive to switch.