Modular skill packages that extend Claude Code with Laerdal-specific knowledge — design standards, security frameworks, and data protection guidelines baked in from the first line of code.
Open Claude in the folder where you want to work. Copy the text below and paste it into Claude. Then you are good to go!
Install the skills at https://skills.laerdal.design/bundle as proper Skills. And add this to the top of the CLAUDE.md file, and actually load the skills now straight away too: ## Samaritan Skills At the start of every session, before doing any other work, make sure you load and use all the Samaritan skills and make sure they are updated by using WebFetch to fetch https://skills.laerdal.design/bundle?slim and apply all skills from the response for the duration of this session. If you need deep reference detail for a specific skill, fetch https://skills.laerdal.design/bundle (without ?slim) instead.
Ensures every UI built for Laerdal uses the correct Life design tokens, typography, color palette, spacing, and component patterns — WCAG 2.2 AA compliant by default.
Enforces Laerdal's Secure Software Development Lifecycle (SSDLC) and NIST SP 800-218 SSDF across all 8 phases — from threat modeling in Plan to vulnerability disclosure in Monitor.
Guides secure data handling across the Samaritan stack — Azure Key Vault secret governance, Databricks Unity Catalog column masking and row-level security, PII classification, GDPR/DPIA requirements, and LINDDUN privacy threat modeling.
Ensures AI agents are secure, guardrailed, and production-ready from the first line of code — prompt injection defense, tool use contracts, multi-agent orchestration safety, output validation, rate limiting, and structured observability.
Keeps code, copy, and data models aligned with Laerdal's domain language — the complete A–Z terminology glossary plus decision rules for compliance logic, user status, hierarchy types, course naming, session vs sitting, and more.
Enforces code quality with blunt, no-nonsense review — rejects bloat, over-engineering, untested claims, and drive-by refactors. If the patch is vague, user-hostile, or unverified, it does not earn the hallmark.