The 20 Claude Code Repositories You Actually Need

Don't want to figure this out alone? I walk members through every step inside the community. Join the Skool → skool.com/raycfu

There are thousands of Claude Code repositories on GitHub right now. Most are forks of forks, half-finished experiments, or repos with a cool README and nothing behind it. I went through the entire ecosystem and narrowed it down to the 20 that actually matter. One or two from each major category. Install these and Claude Code goes from a basic terminal assistant to a full engineering system.

AWESOME LISTS AND INDEXES

These are community-maintained directories that track the entire Claude Code ecosystem. Bookmark them and check back regularly because new repos appear every day.

1. awesome-claude-code (28.5K stars) The definitive curated list of the entire Claude Code ecosystem. Skills, hooks, commands, orchestrators, plugins. Only Claude itself is allowed to submit PRs so the quality stays high. If you only bookmark one thing from this guide, make it this one. Everything you need to discover is indexed here. https://github.com/hesreallyhim/awesome-claude-code
   
2. awesome-claude-skills by ComposioHQ (community maintained) 50+ production-ready Claude skills for automation. PostgreSQL, deep research, CSV analysis, root-cause tracing. These are not toys. They are cross-compatible with Claude Code, Codex, Cursor, and Gemini CLI. Install with /skill add. https://github.com/ComposioHQ/awesome-claude-skills
   

ANTHROPIC REPOS

Start here before you touch anything from the community. These are Anthropic's own projects. Most stable, best documented, most reliable.

1. anthropics/claude-code (55K stars) The official Claude Code terminal tool itself. Natural language commands for your codebase. This is the foundation everything else on this list is built on top of. If you do not have this installed, nothing else matters. https://github.com/anthropics/claude-code
   
2. anthropics/skills (37.5K stars) Official public repository of Agent Skills maintained by Anthropic. PDF, DOCX, XLSX, PPTX creation plus art generation. These are the canonical reference skills that define how skills work in the ecosystem. Start with these before installing community skills so you understand the format. https://github.com/anthropics/skills
   

SKILLS COLLECTIONS

Skills are Markdown files that teach Claude how to perform specific tasks. Good skills turn Claude from a general-purpose assistant into an expert in a specific field on demand.

1. Superpowers by obra (148K stars) The most popular skills library for Claude Code by a massive margin. Gives Claude a full development workflow: brainstorm, spec, plan, TDD, review, merge. It includes subagent orchestration so Claude can delegate parts of a task to specialized sub-agents. Install this first. It changes how Claude approaches every project. https://github.com/obra/superpowers
   
2. everything-claude-code (hackathon winner) Won the Cerebral Valley x Anthropic hackathon. 30 agents, 136 skills, 60 slash commands, 1,282 tests at 98% coverage. Built from 10 months of real daily usage, not a weekend experiment. This is the most complete Claude Code setup that exists. If you want to see what a fully built-out Claude Code environment looks like, study this repo. https://github.com/affaan-m/everything-claude-code
   
3. Karpathy Skills (44K stars in 7 days) One CLAUDE.md file with four principles derived from Andrej Karpathy's public complaints about how AI coding tools behave. Stops Claude from overcomplicating code, touching files you did not ask it to, guessing instead of asking, and failing to verify its own work. Install with one command: /plugin marketplace add forrestchang/andrej-karpathy-skills. Works across every project from that moment forward. https://github.com/forrestchang/andrej-karpathy-skills
   

AGENTS AND SUBAGENTS

A single Claude session cannot handle everything at a high level. Context gets cluttered, files pile up, and by the time you ask for a code review Claude has forgotten what you were building. Subagents solve this by giving you a team of specialists, each with its own context and a single focus.

1. wshobson/agents (25K stars) Massive collection of production subagents covering strategy, development, security, design, data, and research. Drop them into any Claude Code workflow and you have instant domain-specific expertise. The security agents alone are worth the install if you ship anything to production. https://github.com/wshobson/agents
   
2. awesome-claude-agents (3.7K stars) A team of specialized AI agents designed to work together. They collaborate on full features, bringing expert-level knowledge across different tech stacks. Good for understanding how multi-agent collaboration works in practice rather than just in theory. https://github.com/vijaythecoder/awesome-claude-agents
   

MCP SERVERS

On its own, Claude Code reads files, executes commands, and interacts with APIs in your terminal. That is it. It cannot work with GitHub pull requests, query Postgres, open a browser, or read Figma frames. MCP servers act as bridges connecting Claude to the rest of your stack.

1. playwright-mcp (Microsoft official) Official Microsoft browser automation MCP server. Lets Claude navigate pages, fill forms, click buttons, and scrape dynamic content. This is the connector you reach for when a tool or website does not have an API. Claude can interact with any web application through the browser the same way a human would. https://github.com/microsoft/playwright-mcp
   
2. claude-context-mode (2.2K stars) MCP server that sits between Claude Code and your outputs. Achieves 98% reduction in context size (315 KB down to 5.4 KB). The token savings on every single call add up fast. If you are hitting context limits or burning through credits, this is the fix. https://github.com/d-e-s-o/claude-context-mode
   
3. context7 (community maintained) Gives Claude access to up-to-date, version-specific documentation in real time. When Claude writes code using a library, it pulls the current docs for the exact version you are using instead of relying on training data that might be outdated. Works with Cursor, VS Code, Claude Code, and Windsurf. https://github.com/upstash/context7
   

ORCHESTRATION AND WORKFLOWS

These let you run parallel Claude sessions, coordinate multi-step workflows, preserve state between runs, and manage agents like a small team.

1. claude-squad (5.6K stars) Terminal multiplexer for AI coding agents. Manages Claude Code, Aider, Codex, OpenCode, and Amp in parallel sessions. One agent builds a feature, another writes tests, a third refactors old code. All running at the same time without interfering with each other. This is the tool that turns Claude from a single worker into a team you manage. https://github.com/smtg-ai/claude-squad
   
2. Claude-Flow (11.4K stars) Enterprise-grade AI orchestration platform. Multi-agent orchestration with persistent memory. If claude-squad is managing a few parallel sessions, Claude-Flow is managing an entire engineering department. More setup required but significantly more powerful for complex multi-agent workflows. https://github.com/ruvnet/claude-flow
   

MEMORY AND CONTEXT

Claude Code's biggest drawback is that every session starts from scratch. These repos give Claude persistent memory so it stops forgetting your architecture, your preferences, and your past decisions.

1. claude-subconscious (2.4K stars) A background agent that watches your sessions, reads your files, and builds memory over time. It runs quietly alongside your normal Claude sessions and accumulates knowledge about your projects. Over time Claude starts remembering things you never explicitly told it because the subconscious agent picked them up. https://github.com/0xfurai/claude-subconscious
   
2. Repomix (20.9K stars) Packs your entire codebase into one AI-readable file. Outputs as XML, Markdown, or plain text. Essential for large repos where Claude keeps losing context because it cannot see the full picture. You feed Repomix your project and it produces a single file that gives Claude complete visibility into the codebase structure. https://github.com/yamadashy/repomix
   

SLASH COMMANDS AND HOOKS

Slash commands replace repetitive multi-sentence instructions with a single keystroke. Hooks trigger automatically at key moments in your workflow so you do not have to remember to enforce your own rules.

1. wshobson/commands (1.7K stars) 57 production-ready slash commands covering 15 workflows and 42 tools. Multi-agent orchestration built in. Instead of typing "review this code for security issues, check for performance problems, and suggest refactors" you type /review and it handles everything. The time saved per day compounds fast. https://github.com/wshobson/commands
   
2. tdd-guard (1.7K stars) Automated TDD enforcement for Claude Code. If Claude tries to commit code without writing tests first, tdd-guard blocks it and explains why. It does not just remind you to write tests. It prevents Claude from skipping them entirely. This is the kind of guardrail that makes the difference between a demo and production-quality code. https://github.com/nizos/tdd-guard
   

GUIDES AND LEARNING

Claude Code has hidden complexities that nobody talks about. Prompt development, skill creation, MCP configuration, hook design, context management, and token economics. These guides cover what would take months to figure out on your own.

1. claude-prompt-engineering-guide (community maintained) The definitive prompt engineering guide for Claude Code. Covers Opus 4.6, Sonnet 4.6, and Haiku 4.5. Includes sections on Skills, Plugins, MCP, Hooks, and Ultrathink. Backed by 220+ verified sources. If you want to understand how to get the best output from Claude Code, this is the single most thorough resource available. https://github.com/ThamJiaHe/claude-prompt-engineering-guide
   
2. ClaudeWave (real-time directory) A real-time directory tracking 3,800+ Claude Code repositories across 8 categories. Tagged by use case, ranked by stars and velocity, updated every 12 hours. Think of it as a live dashboard of the entire Claude Code ecosystem. When something new blows up on GitHub, ClaudeWave catches it within hours. Good for staying current without manually scrolling GitHub trending pages. https://github.com/claudewave/claudewave
   

HOW TO INSTALL

Different repos install differently depending on what they are.

For skills: /skill add [repo-url] or drop the Markdown file into your project.

For plugins: /plugin marketplace add [author/repo] then /plugin install [plugin-name].

For CLAUDE.md files: Copy the file into your project root. Claude reads it automatically at the start of every session.

For MCP servers: Add the server config to ~/.claude/config.json under the mcpServers section. Each repo's README has the exact JSON to paste.

For slash commands: Drop the command files into your .claude/commands/ directory.

For subagents: Most install via /skill add or by copying the agent files into your project.

Check each repo's README for the exact install command. Most take under 60 seconds.

WHERE TO START

If you install nothing else, install these three today:

Superpowers for the full development workflow. This changes how Claude approaches every single task.

The Karpathy skills file for the behavior fixes. This stops Claude from overcomplicating, guessing, and touching things it should not.

Repomix for codebase visibility. Feed Claude your entire project in one file so it stops losing context.

Those three alone will make Claude Code feel like a completely different tool. Add the rest as you need them. The repos are not going anywhere.

You just read the full playbook. Most people will close this tab and never implement it. The ones who do usually hit a wall around the technical setup and quit.

Inside the Skool, I walk you through the exact build step-by-step, troubleshoot your setup live in the community, and share the scripts and templates I use to actually land paying clients.

If you want the shortcut instead of the long way around:

Join the Skool →