Skill Grafting
Call windags_skill_graft when your task requires domain expertise you don't have in context.
When to Graft
- Architecture decisions (caching strategies, API design, database schema)
- Framework-specific work (React Server Components, SwiftUI data flow, Terraform modules)
- DevOps and deployment (Docker multi-stage builds, Kubernetes manifests, CI/CD pipelines)
- Security implementation (OAuth flows, RBAC, encryption patterns)
- Specialized domains (data pipelines, ML model serving, real-time collaboration)
When NOT to Graft
- Simple code changes (rename, format, delete unused code)
- Tasks you can complete confidently from general knowledge
- When you've already grafted skills for this task
How to Use
- Call
windags_skill_graftwith a clear task description and count (usually 2) - Read the returned skill bodies — they contain decision trees, failure modes, and worked examples
- Check the
referenceSummaries— these summarize deeper material available viawindags_skill_reference - If you need the full content of a reference, call
windags_skill_reference(skillId, path) - Apply the skill's quality gates to validate your work