Claude Code Best Practices

Summary of: deepresearch/dev-workflow/claude-code-best-practices.md

Key Points

• CLAUDE.md structure: Include project overview, tech stack, architecture, coding standards, and AI-specific guidelines
• Keep CLAUDE.md focused: Under 500 lines; avoid including entire codebases or irrelevant details
• Custom commands: Create .claude/commands/ for repetitive tasks; use frontmatter for model, tools, description
• MCP (Model Context Protocol): Configure servers for external tool access (databases, APIs, file systems)
• Allowed tools: Restrict tool access per command for security and token efficiency
• Extended thinking: Use "think", "think hard", "ultrathink" keywords to increase reasoning depth
• Hooks system: Configure pre/post hooks for tool calls (validation, logging, notifications)
• Session management: Use /compact to condense context, /clear between unrelated tasks

Critical Insights

1. Context quality > context quantity - Focused, well-structured context outperforms verbose context
2. Command library is force multiplier - Well-designed commands encode team knowledge and reduce repetitive prompting
3. Tool restrictions serve dual purpose - Both security (limit damage) and efficiency (reduce token overhead)

Quick Reference

Feature	Location	Purpose
CLAUDE.md	Project root	Project context for all sessions
Commands	.claude/commands/	Reusable task templates
Agents	.claude/agents/	Specialized subagent configurations
Settings	.claude/settings.json	Global/project settings
Hooks	Settings file	Pre/post tool call automation

CLAUDE.md Template Sections

1. Project Overview
2. Tech Stack & Dependencies
3. Architecture Patterns
4. Coding Standards
5. Testing Guidelines
6. AI Behavioral Guidelines
7. What NOT to do

Command Frontmatter Options