Improving HCLS AI reasoning with open-source agent skills
aws ·
AWS has released an open-source collection of 38 agent skills designed to improve AI reasoning in healthcare, life sciences, and related fields. These skills address the gap where foundation models misapply complex decision frameworks despite having access to relevant guidelines. By encoding structured methodologies into markdown documents, the skills guide agents through precise procedures for tasks like variant interpretation and claims adjudication. Evaluation results show that agents equipped with these skills win 70–86% of head-to-head comparisons against unskilled counterparts. The most significant improvements were observed in critical thinking, with Cohen’s d effect sizes ranging from 0.65 to 1.03. The solution supports multiple deployment patterns, including single-agent setups in Quick Desktop and multi-agent orchestration via Kiro CLI. Additionally, the skills are portable across various AWS services and third-party coding agents, allowing for easy customization and extension by users.
aws