Claude-Code-Everything-You-Need-to-Know

Enables developers to define reusable AI-assisted workflows as markdown files stored in .claude/commands/ directory. Each skill file contains prompts, instructions, and context that Claude executes when invoked via /skillname syntax. The system parses markdown metadata to extract skill definitions and automatically registers them as CLI commands, allowing non-programmers to extend Claude Code's capabilities without writing code.

Maintains a CLAUDE.md file in the project root that stores persistent context, decisions, architecture notes, and project state. This file is automatically parsed and injected into every Claude interaction, eliminating the need to re-explain project context. The system treats CLAUDE.md as a living document that Claude can read and suggest updates to, creating a feedback loop where project knowledge accumulates across sessions.

Maintains session state across multiple CLI invocations, preserving conversation history, variable bindings, and execution context. Developers can continue conversations across separate claude commands without re-explaining context. Sessions are stored locally and can be resumed, forked, or archived, enabling complex multi-step workflows to be broken into manageable CLI invocations while maintaining continuity.

Provides slash commands (/init, /model, /fast, /help, etc.) for core operations like project initialization, model selection, fast mode toggling, and help. Commands are implemented as built-in handlers in the CLI process and execute immediately without invoking Claude. The command interface is extensible; custom skills can be invoked as commands, creating a unified command namespace for both system operations and user-defined workflows.

Enables agents to spawn subagents to handle subtasks, creating hierarchical task decomposition. Parent agents can define subtasks, delegate to subagents, and aggregate results. Subagents inherit parent context (CLAUDE.md, project memory) but can have specialized prompts and tool bindings. This pattern enables complex problems to be solved through recursive decomposition without requiring manual task management.

Provides experimental support for agent teams that collaborate on shared tasks using communication patterns like voting, consensus-building, and debate. Multiple agents with different perspectives or specializations work together to solve a problem, with a coordinator agent aggregating results and resolving disagreements. This enables more robust solutions by leveraging diverse viewpoints and reducing single-agent errors.

Enables spawning multiple AI agents that work in parallel on different branches using git worktrees. Each agent operates in an isolated working directory, executes tasks independently, and reports results back to a coordinator. The system manages branch creation, agent lifecycle, and result aggregation, allowing complex development tasks to be decomposed and executed concurrently by specialized agents (e.g., frontend, backend, database agents).

Provides pre-configured agent templates (Business Analyst, Project Manager, UX Engineer, Database Engineer, Frontend Engineer, Backend Engineer, Code Reviewer, Security Reviewer) that encapsulate role-specific prompts, tools, and decision-making patterns. Each template is instantiated as an agent with specialized context and MCP server bindings, enabling developers to delegate work to agents that understand domain-specific concerns and can operate autonomously within their expertise area.

Provides /model command to switch between Claude Opus 4.6 (most capable), Sonnet 4.5 (balanced), and Haiku 4.5 (fastest/cheapest) models, and /fast toggle to enable fast mode which uses Haiku for initial reasoning then escalates to selected model if needed. The system tracks token usage across sessions and provides recommendations for model selection based on task complexity and budget constraints, enabling developers to optimize cost-performance tradeoffs.

Integrates MCP servers as external capability extensions configured in ~/.claude.json. Includes built-in Serena semantic code intelligence server that provides AST-based code analysis, symbol resolution, and codebase-aware context. Developers can register custom MCP servers for domain-specific tools (e.g., database clients, API integrations, security scanners), and Claude automatically discovers and uses available tools based on task context.

Enables Claude to break down complex problems into sequential reasoning steps using extended thinking patterns. Developers can invoke sequential thinking mode to have Claude explicitly decompose tasks, explore multiple solution paths, and explain its reasoning before providing final answers. This is particularly useful for architectural decisions, debugging complex issues, and algorithm design where transparency and correctness are critical.

Provides a hooks system that triggers custom skills or agents at specific lifecycle points (e.g., pre-commit, post-merge, on-file-change). Hooks are defined in CLAUDE.md or .claude/hooks/ and execute automatically when conditions are met, enabling developers to automate code quality checks, documentation updates, or deployment steps without manual invocation. The system integrates with git hooks and file watchers to detect trigger conditions.

Integrates Playwright as an MCP server to enable Claude to automate browser interactions, perform web testing, and extract data from web applications. Claude can write and execute Playwright scripts to navigate pages, fill forms, take screenshots, and validate UI behavior. This enables end-to-end testing, web scraping, and UI validation workflows to be automated and integrated into development pipelines.

Centralizes all Claude Code configuration in ~/.claude.json, including API keys, model preferences, MCP server definitions, and user settings. The configuration file uses JSON schema with support for environment variable interpolation, enabling developers to manage settings across machines and share configurations with teams (minus sensitive credentials). Changes to ~/.claude.json are applied immediately without restarting the CLI.