DSS-Skills: A Decision-Workflow Taxonomy of Reusable Procedural Capabilities for Decision Support Systems

Abstract

Decision support systems (DSS) increasingly combine data, models, rules, workflows, and agentic tool use, but they still offer limited support for establishing the defensibility of decision outputs: framing the problem, qualifying evidence, bounding claims, generating options, comparing alternatives, and learning from outcomes. Existing agent-skill researches treat skills as reusable procedural units, but mainly classifies them by lifecycle concerns such as representation, acquisition, retrieval, selection, and evolution. This leaves a decision-function localization gap: once a skill-like procedure is embedded in a DSS, it remains unclear which part of decision-support work it externalizes and what kind of decision artifact it assumes. This paper defines DSS-skills as reusable procedural capabilities that execute recurring decision-support work under defined decision contexts, evidence requirements, judgment criteria, authority boundaries, and feedback conditions. Based on 45 core studies describing reusable procedures, workflows, agent routines, or skill-like mechanisms relevant to decision support, we develop a decision-workflow taxonomy with six primary categories and 24 functional subtypes: problem-framing, situation-sensing and evidence-integration, analytic-inference, option-generation, evaluation-and-deliberation, and execution-monitoring and learning skills. The taxonomy classifies DSS-skills by their primary decision-support function. It also highlights three cross-category challenges essential for building organizational capabilities: bounded triggering for reuse, workflow-level validity for composition, and governed feedback for revision. This framework helps design skill templates, organize workflows, and manage feedback in agent-based decision systems.