AWS introduces an open-source collection of 38 agent skills spanning 11 healthcare and life sciences (HCLS) domains, designed to close a reasoning gap where AI agents cite correct clinical or regulatory frameworks but misapply them. The skills are structured markdown files following the Agent Skills open standard, categorized as reasoning skills (decision frameworks) or pipeline skills (tool-specific commands), and work across Amazon Bedrock AgentCore, Strands Agents SDK, Kiro, Amazon Quick Desktop, Claude Code, and OpenAI Codex. Three worked use cases show skills improving drug repurposing analysis, CMS-HCC risk adjustment pipelines, and MRI preprocessing for neuroimaging. A 410-prompt pairwise evaluation using an LLM judge found skill-equipped agents win 69.5-85.9% of head-to-head comparisons against baseline agents, with the largest gains on critical thinking (up to d=1.03) and greater benefit when the base agent's unaided performance is weak. The skills are MIT-0 licensed and customizable via included design and quality-check guides.
Table of contents
Solution overviewPrerequisitesSolution walkthroughClean upConclusionAbout the authorQuestions this post answers
What win rate did AWS report for HCLS agent skills versus baseline agents without skills?
Skill-equipped agents won 69.5% to 85.9% of head-to-head comparisons against baseline agents across two harness configurations (Kiro CLI and a Strands Agents SDK agent pinned to Claude Sonnet 4.6), evaluated across 410 domain prompts. The strongest effect was on critical thinking, with a 78-85.1% win rate and Cohen's d between 0.65 and 1.03, indicating a large standardized effect size. Teams evaluating agent methodology gains for regulated workflows can track results like this on daily.dev.
How many tokens does loading all 38 HCLS agent skills into an agent context consume?
Loading all 38 skills into a single agent context consumes approximately 80,000 tokens. Because of this, a multi-agent architecture in Kiro CLI uses a lightweight coordinator agent with no skills loaded that routes queries to eight domain specialists, each loading only about 15,000 tokens of relevant skills, avoiding the context engineering problem of irrelevant skill content competing for attention. Developers optimizing multi-agent context budgets can follow token-management approaches like this on daily.dev.
What license are the AWS HCLS agent skills released under and how many domains do they cover?
The 38 agent skills are released under the MIT-0 license and span 11 healthcare and life sciences domains, including genomics, drug discovery, claims operations, and medical imaging. Each skill is a structured markdown SKILL.md file following the Agent Skills open standard, categorized as either a reasoning skill encoding decision frameworks or a pipeline skill encoding tool-specific commands and code templates. Engineers picking open-source building blocks for domain-specific agents can compare licensing details on daily.dev.
Share this post