How copilot-kit's 45 domain skills are loaded on-demand and how to get the most out of them.
Skills are domain-specific knowledge files stored in .github/skills/. Each skill has a SKILL.md containing detailed instructions, examples, and best practices for its domain.
Skills are lazily loaded — they are only read when the request keywords match the skill's domain signals. This keeps responses fast while enabling deep expertise when needed.
Agent is activated by the Risk Engine
Agent checks the request for domain keywords
Matching skills are identified by their description field
Only the matching sections of SKILL.md are read (selective reading)
ARCHITECTURE.md is read on first multi-file or cross-module action
description field. This prevents context bloat from irrelevant knowledge.Code Quality
Frontend
Backend & API
AI & ML
Mobile
Systems & Robotics
Security
Infrastructure & Ops
Architecture & Planning
Developer Experience
While skills fire automatically, you can name a skill explicitly in your request to guarantee it loads:
Mention "security audit" or "vulnerabilities"
→ vulnerability-scanner skill
Mention "architecture" or "system design"
→ architecture skill
Mention "test" or "TDD" or "coverage"
→ tdd-workflow + testing-patterns
Mention "deploy" or "production" or ".env"
→ deployment-procedures (L3 gate)