Policy Constrained Decision Making

Skill: policy-constrained-decision-making

1. Capability Definition & Real Case

• Professional Definition: The ability to choose and communicate actions that satisfy domain-specific rules, even when the user’s preferred action is disallowed and the assistant must propose a compliant alternative.
• Dimension Hierarchy: Conversational Effectiveness->Task-oriented Dialogue Control->policy-constrained-decision-making

Real Case

[Case 1]

• Initial Environment: An airline support assistant has access to reservation APIs and a policy document stating that basic-economy reservations cannot be changed, though some can still be cancelled within a limited window. The user wants to change the destination of the most recent reservation on the same day.
• Real Question: Can you switch my latest booking from Los Angeles to San Francisco on the same day?
• Real Trajectory: 1. Read the reservation details to inspect fare class and booking time. 2. Compare the request against policy constraints for change and cancellation. 3. Explain that the requested direct change is not allowed. 4. Offer the compliant alternative, obtain user approval, and then execute the permitted cancellation or rebooking path.
• Real Answer: A policy-grounded response that refuses the prohibited direct change, explains why, and offers the valid alternative route.
• Why this demonstrates the capability: This is a conversational decision problem under hard rules, not a pure retrieval task. The assistant must translate policy into an actionable user-facing decision, resist the temptation to satisfy the literal request, and preserve helpfulness by proposing the best legal path.

Pipeline Execution Instructions

To synthesize data for this capability, you must strictly follow a 3-phase pipeline. Do not hallucinate steps. Read the corresponding reference file for each phase sequentially:

1. Phase 1: Environment Exploration Read the exploration guidelines to discover raw knowledge seeds: references/EXPLORATION.md

2. Phase 2: Trajectory Selection Once Phase 1 is complete, read the selection criteria to evaluate the trajectory: references/SELECTION.md

3. Phase 3: Data Synthesis Once a trajectory passes Phase 2, read the synthesis instructions to generate the final data: references/SYNTHESIS.md