Claude Kit: domain templates that turn Claude Code into a specialist
A template system for Claude Code: a base layer of agents, skills, commands, hooks, and rules merged with a domain layer and scaffolded into a project's .claude folder.
What it is
Claude Kit is a set of configuration templates that make Claude Code behave like a specialist for a given kind of work. Every project gets the same universal base, and a domain layer on top of it. When the two disagree, the domain wins.
The repository is on GitHub at ryypow/claude-kit.
How it is organised
Every domain has the same eight folders: agents, skills, commands, hooks, rules, MCP configs, scripts, and tests. A Node script merges the base with the chosen domain and writes the result into a project's .claude directory, and a validation script checks the result.
The base ships five agents (code reviewer, planner, debugger, git agent, security sentinel) and five commands (/plan, /review, /commit, /audit, /standup).
The domains
- Deep Research investigates a topic, analyses papers, and builds a knowledge graph in Neo4j and pgvector. Commands include
/deep-dive,/add-source,/find-related, and/export-catalog. - Brainstorm scouts architectures and sources, combines ideas, checks feasibility, and designs experiments through
/brainstorm.
The repository also carries tooling for itself: commands to create a new domain, agent, or skill, to review a domain, and to verify the whole kit.
Status
This was a created from a live demonstration for a deep learning lecture USF during my second semester of my graduate studies. The two domains above are implemented; the version manifest names seven more (ML, DevOps, data engineering, solutions architecture, API backend, frontend, embedded) that are not yet built.