Summary
Highlights
Claude Code offers several customization options like Skills, Claude.md, Sub-agents, Hooks, and MCP servers. Each solves different problems, and understanding their distinct uses prevents incorrect implementation.
Claude.md loads into every conversation, always. Use it for project-wide standards, constraints (e.g., never modify database schema), framework preferences, and coding style. Skills load on demand when Claude matches a request. They are for task-specific expertise, knowledge relevant only sometimes, and detailed procedures that would otherwise clutter every conversation.
Hooks fire on events, such as running a linter every time Claude saves a file or validating input before tool calls. They are event-driven. Skills are request-driven, activating based on what you're asking. Use hooks for operations that run on every file save, validations before specific tool calls, or automated side effects. Use skills for knowledge that informs how Claude handles requests and guidelines affecting Claude's reasoning.
A typical setup involves Claude.md for always-on project standards, skills for task-specific expertise, and hooks for automated operations. Each feature has its specialty, and it's important not to force everything into skills when another option fits better. Multiple options can be used simultaneously for comprehensive customization.
Skills provide automatic task-specific expertise. Claude.md is for always-on instructions. Sub-agents run in isolated contexts. Hooks fire on events. MCP provides external tools. Combine these features for comprehensive and effective Claude Code customization.
Skills add knowledge to your current conversation, with instructions joining the existing context. Sub-agents run in a separate, isolated context, receiving a task, working independently, and returning results. Use sub-agents to delegate tasks that require a separate execution context, different tool access, or isolation from the main conversation. Use skills to enhance Claude's knowledge for the current task where expertise applies throughout a conversation.