Claude Code: All Features For Power Users

Key Concepts

• Claude Code: An agentic command-line interface (CLI) tool that interacts with your local file system to perform coding tasks.
• MCP (Model Context Protocol): A standard for connecting AI agents to external data sources and APIs.
• Claude.md: Markdown files used to provide persistent project-specific instructions and guidelines.
• Skills: Modular, fragmented instruction files that allow Claude to load specific context only when needed.
• Sub-agents: Specialized AI agents spawned by a master agent to handle specific tasks (e.g., UI design vs. backend).
• Agent Teams: An experimental feature where multiple agents communicate in a decentralized manner to solve complex problems.
• Git Worktrees: A Git feature allowing multiple branches of a repository to be checked out into separate directories simultaneously.

1. Installation and Basic Usage

Claude Code functions as an autonomous engineer on your local machine. It can read, write, and execute commands with user permission.

• Installation: Available via package managers (e.g., AUR for Arch Linux) or as a standalone application.
• Authentication: Requires a subscription; users authenticate via the terminal.
• Workflow: Users can initiate projects (e.g., uv init) and interact via natural language. The agent requests permission for file edits, which can be granted individually or globally for a session.

2. Essential Slash Commands

• /model: Switches between models. Claude 3.5 Sonnet is recommended for speed/cost, while Claude 3.5 Opus is preferred for complex refactoring and large codebases.
• /rewind: An undo function that reverts code changes and conversation history to a previous state.
• /resume: Allows users to pick up previous sessions by selecting from a list of identifiers.
• /usage & /stats: Tracks token consumption, session limits, and provides insights into productivity patterns.
• /config: Adjusts settings like "fast mode," "thinking mode," and verbosity.
• /insights: Generates an HTML report analyzing the user's workflow, identifying bottlenecks, and providing statistics on tool usage and error rates.

3. Context Management: Claude.md vs. Skills

• Claude.md: Used for global or project-wide instructions (e.g., coding style, project structure, or specific tool preferences). It is always loaded into context.
• Skills: A modular alternative to claude.md. By placing instructions in .claude/skills/[skill_name]/skill.md, Claude only loads the relevant instructions when the task requires them, saving context window space.
  • Methodology: Each skill file must contain a header with a name and description to allow the agent to trigger it appropriately.

4. Extending Functionality: MCP and Plugins

• MCP (Model Context Protocol): Allows Claude to interact with external APIs (e.g., Google Calendar, Deep Wiki).
  • Command: claude mcp add --transport [http/stdio] [server_url]
• Plugins: A marketplace-based system that bundles skills and MCP servers.
  • Process: Use /plugins to discover and install tools. After installation, run reload plugins to apply changes to the current session.

5. Advanced Agentic Architectures

• Sub-agents: The master agent delegates tasks to specialized agents.
  • Process: Define agents with specific roles (e.g., "UI Designer" vs. "Backend Specialist"). You can restrict tool access per agent and assign different models (e.g., Sonnet for UI, Opus for Backend).
• Agent Teams: An experimental feature (enabled via CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1) where agents communicate with each other in a decentralized way rather than reporting back to a single master. This is useful for brainstorming or multi-angle problem solving but is token-intensive.

6. Git Worktrees and Memory

• Worktrees: By using claude --worktree, users can spawn multiple instances of Claude Code working on different branches of the same repository in separate directories. This is ideal for A/B testing or parallel feature development.
• Memory: Claude can store persistent facts across sessions. By instructing the agent to "remember" specific details (e.g., user name or project preferences), it will recall this information in future sessions.

Synthesis

Claude Code is a powerful tool for power users who want to move beyond simple chat-based coding. By leveraging Skills for modular context, MCP for external data integration, and Sub-agents for specialized task delegation, developers can create a highly efficient, automated coding environment. While features like Agent Teams remain experimental, the combination of Git Worktrees and persistent Memory provides a robust framework for managing complex, multi-faceted software projects.