# If You Are Not Using These 100 Repositories, You Are Using Claude Code Wrong.
**作者**: NeilXbt
**日期**: 2026-05-18T14:48:08.000Z
**来源**: [https://x.com/neil_xbt/status/2056386359376396458](https://x.com/neil_xbt/status/2056386359376396458)
---
Most people think of Claude Code as a smart terminal assistant.
You type something. It writes code. You review it. You move on. Useful, sure. But not fundamentally different from what you were doing before, just faster.
That was how I thought about it too. Until I discovered that the real Claude Code, the version that other developers are building entire companies with, the version that runs 30 parallel agents, maintains memory across sessions, enforces TDD automatically, integrates with every tool in your stack, and compounds in capability every week, is not the one you get out of the box.
It is the one you assemble from repositories.
The ecosystem around Claude Code has grown faster than almost anyone has tracked. New skills, plugins, subagents, MCP servers, orchestrators, memory systems, IDE integrations, and workflow harnesses are appearing daily. Most of them never make it to Hacker News. Most of them never get the stars they deserve. And most Claude Code users have never heard of any of them.
That gap, between what Claude Code is by default and what it becomes when you start building on top of it, is exactly where the productivity difference lives. The developers who are 5x more effective with the same subscription are not smarter. They are not working harder. They installed a few repositories and built a system instead of using a tool.
This article covers 100 of the best repositories across 11 categories. Every one of them is real, maintained, and immediately usable. By the end, you will know exactly what to install and why.
## Category 1: Awesome Lists and Meta-Indexes
Before you install anything specific, you need a map of the territory. The Claude Code ecosystem is evolving faster than any single person can track. Awesome lists solve this by crowdsourcing the curation, filtering by actual quality rather than GitHub stars, and keeping pace with an ecosystem that shifts weekly.
Start here. Come back here often.
1. awesome-claude-code
The definitive curated list of the entire Claude Code ecosystem. This is the main index with 28,500 stars and a remarkable quality control mechanism: only Claude itself is allowed to submit pull requests. Skills, hooks, commands, orchestrators, and plugins are all here, filtered for genuine usefulness rather than just presence. repo: https://github.com/hesreallyhim/awesome-claude-code
2. awesome-claude-skills (ComposioHQ)
Over 50 production-ready Claude skills for automation, covering PostgreSQL, deep research, CSV processing, and root-cause tracing. Cross-compatible with Claude Code, Codex, Cursor, and Gemini CLI. Install directly via /skill add. repo: https://github.com/ComposioHQ/awesome-claude-skills
3. awesome-claude-plugins (ComposioHQ)
A curated plugin collection covering commands, agents, hooks, and MCP servers. Includes connect-apps, frontend-design, and artifacts-builder. Install via /plugin marketplace add. repo: https://github.com/ComposioHQ/awesome-claude-plugins
4. awesome-claude-prompts
A 4,200 star collection of optimized Claude prompts using XML-structured prompt techniques. Covers business, creative, and development workflows and is copy-paste ready from day one. repo: https://github.com/langgptai/awesome-claude-prompts
5. awesome-claude
An all-around list of everything Anthropic Claude-related: courses, guides, desktop apps, articles, MCP coverage, the BMAD method, and Claude Code tips. The best starting point for newcomers who want to understand the full landscape before diving into specifics. repo: https://github.com/alvinunreal/awesome-claude
6. awesome-agent-skills (VoltAgent)
Over 1,000 agent skills sourced from real engineering teams rather than AI generation. Battle-tested and compatible with Claude Code, Codex, Gemini CLI, and Cursor. Includes Notion, WhatsApp, NotebookLM, and memory skills. repo: https://github.com/VoltAgent/awesome-agent-skills
7. awesome-claude-code-subagents (VoltAgent)
Hand-picked subagents for every stack and task, organized into domain-specific expertise bundles. Production-ready components that integrate with most coding agents. repo: https://github.com/VoltAgent/awesome-claude-code-subagents
8. awesome-claude-code-plugins (ccplugins)
Focused specifically on Claude Code plugins: slash commands, subagents, MCP servers, and hooks, with install commands included for each entry. Also enables hosting your own marketplace. repo: https://github.com/ccplugins/awesome-claude-code-plugins
9. awesome-claude-code (jqueryscript)
A curated collection of tools, IDE integrations, and frameworks with star-tier markers. Covers SDKs, security tools, and agent collections. Excellent as a discovery tool for the less-visible parts of the ecosystem. repo: https://github.com/jqueryscript/awesome-claude-code
10. Claude-Code-Repos-Index
An indexed catalogue of 75+ quality Claude Code repositories, all original with no AI-generated filler, maintained by Daniel Rosehill. A practical navigator for the ecosystem with a focus on daily usability rather than novelty. repo: https://github.com/danielrosehill/Claude-Code-Repos-Index
## Category 2: Official Anthropic Repositories
Before any community repository, start with these. These are Anthropic's official projects: the CLI itself, the SDKs, the skills library, and the GitHub Action. Every community repository on this list is built on top of them.
They may not always have the widest feature range, but they are the most stable, best-documented, and most reliable choice for teams who need something that will not change out from under them.
11. anthropics/claude-code
The official Claude Code terminal tool with 55,000 stars. Natural language commands for your codebase, maintained directly by Anthropic. If you are using Claude Code, you are already using this. But the repository itself contains documentation, issue tracking, and official guidance that most users never read. repo: https://github.com/anthropics/claude-code
12. anthropics/skills
The official public repository of Agent Skills with 37,500 stars. Covers PDF, DOCX, XLSX, PPTX, and art generation. This is the canonical skills reference and the place to start understanding what a well-structured skill actually looks like before you start writing your own. repo: https://github.com/anthropics/skills
13. anthropics/claude-plugins
The Anthropic-managed directory of quality plugins with 2,800 stars. A vetted collection that serves as a reliable baseline for teams who need to know that what they install has been reviewed by people who built the tool. repo: https://github.com/anthropics/claude-plugins
14. anthropics/claude-code-sdk-python
The official Python SDK for Claude Code with 6,100 stars. Provides programmatic access to capabilities and supports Agent SDK patterns. Essential for anyone building automations or integrations on top of Claude Code. repo: https://github.com/anthropics/claude-code-sdk-python
15. anthropics/claude-agent-sdk-typescript
The official TypeScript SDK for Claude Agent with type-safe tool orchestration, streaming support, MCP integration, and a V2 preview featuring send() and stream() patterns. The foundation for any team building production TypeScript tooling on Claude. repo: https://github.com/anthropics/claude-agent-sdk-typescript
16. anthropics/claude-code-security-review
An AI-powered security review GitHub Action with 2,800 stars. Analyzes pull request diffs for vulnerabilities and drops cleanly into existing CI/CD pipelines. One of the most immediately practical repositories on this entire list. repo: https://github.com/anthropics/claude-code-security-review
17. anthropics/claude-code-action
The official GitHub Action for Claude in pull requests. @claude mentions trigger analysis, diff reviews, and patch suggestions, all following your repository's own conventions. repo: https://github.com/anthropics/claude-code-action
18. github/github-mcp-server
The official GitHub MCP server. Provides direct access to repositories, pull requests, issues, and CI/CD with a read-only mode via X-MCP-Readonly. The most widely used server in the entire MCP ecosystem. repo: https://github.com/github/github-mcp-server
19. anthropic-cookbook
Official Anthropic cookbook with working recipes: notebooks, agent patterns, tool-use examples, RAG, classification, and summarization. Maintained by the Anthropic team and the most trustworthy source for seeing what good API usage looks like. repo: https://github.com/anthropics/anthropic-cookbook
20. courses (Anthropic)
Free official Anthropic courses covering prompt engineering fundamentals, API usage, agents, and AI fluency. Direct from the source and the fastest way to build a correct mental model before you start experimenting with community tooling. repo: https://github.com/anthropics/courses
## Category 3: Skills Collections
Skills are the fastest way to expand Claude Code's capabilities without writing a single line of code. They are Markdown files that teach Claude how to perform specific tasks: writing pull requests, creating videos in Remotion, conducting security audits, generating diagrams, running SEO analyses.
The power of a good skill is that it transforms Claude from a general-purpose assistant into a domain expert on demand. You do not explain your tech stack from scratch every session. The skill already knows it.
21. obra/superpowers
The most popular skills library for Claude Code with 148,000 stars. Covers the full development cycle: brainstorm, spec, plan, TDD, review, and merge, with subagent orchestration built in. If you install one skills library, this is the one. repo: https://github.com/obra/superpowers
22. obra/superpowers-marketplace
The official marketplace for the Superpowers ecosystem with over 20 battle-tested skills. Includes /brainstorm, /write-plan, and /execute-plan with SessionStart context injection for seamless workflow continuity. repo: https://github.com/obra/superpowers-marketplace
23. obra/superpowers-skills
The community-editable skills library for Superpowers, auto-cloned to ~/.config/superpowers/skills/. Fork it, submit PRs, and contribute to the skills that the entire Superpowers community uses. The collaborative backbone of the ecosystem. repo: https://github.com/obra/superpowers-skills
24. obra/superpowers-lab
Experimental skills under active development, including semantic duplicate-function detection and bleeding-edge techniques. This is where the next version of Superpowers comes from. Worth watching even if you are not ready to use experimental features. repo: https://github.com/obra/superpowers-lab
25. everything-claude-code
A Cerebral Valley x Anthropic hackathon winner built from 10+ months of real use. Contains 30 agents, 136 skills, 60 slash commands, and 1,282 tests at 98% coverage. One of the most comprehensive single repositories on this list. repo: https://github.com/affaan-m/everything-claude-code
26. travisvn/awesome-claude-skills
A community best-of skills collection with 22,000+ installs covering SEO, marketing, design, security, and research. Includes safety warnings on untrusted skills, which is exactly the kind of curation discipline that makes a community collection trustworthy.
repo: https://github.com/travisvn/awesome-claude-skills
27. trailofbits/claude-code-skills
Security research and audit skills from Trail of Bits, one of the most respected security engineering firms in the industry. If you are working on anything where vulnerability detection matters, this collection was built for exactly that.
repo: https://github.com/trailofbits/claude-code-skills
28. K-Dense-AI/claude-scientific-skills
Skills designed specifically for scientific computing, with specialized libraries for research workflows, academic tasks, and data analysis. The gap in most skill collections is science and research. This fills it. repo: https://github.com/K-Dense-AI/claude-scientific-skills
29. ay-automate/ay-skills
Ten production-tested skills covering SEO audits, headless browser automation, Remotion video generation, and diagrams. Drop-in ready and focused on tasks that most skill collections do not touch. repo: https://github.com/ay-automate/ay-skills
30. BehiSecc/awesome-claude-skills
A meta-resource that functions as a list of lists, covering knowledge acquisition, deeper thinking frameworks, and skill automation utilities. Most useful for people who have already explored the major collections and want to go further. repo: https://github.com/BehiSecc/awesome-claude-skills
## Category 4: Agents and Subagents
A single Claude session has limits. As context grows, performance degrades. You end up with a single agent that is half-aware of too many things rather than a set of specialists who each do one thing well.
Subagents solve this structurally. Instead of one Claude struggling with twenty tasks, you get a team: each agent with its own context window, its own domain expertise, and a single clear responsibility. The orchestrator routes tasks. The specialists execute them.
31. wshobson/agents
A massive collection of production subagents with 25,000 stars, organized across strategy, development, security, design, data, and research domains. Drop-in compatible with Claude Code workflows. One of the most starred and most used repositories on this list. repo: https://github.com/wshobson/agents
32. awesome-claude-agents
A team of specialized AI agents designed to collaborate on complete features. 3,700 stars. Agents are built with expert-level knowledge across different technology stacks and are designed to work together rather than in isolation. repo: https://github.com/vijaythecoder/awesome-claude-agents
33. davepoon/claude-code-subagents-collection
Over 100 production-ready subagents with 2,200 stars and comprehensive domain expertise. Built for drop-in integration with any Claude Code workflow. repo: https://github.com/davepoon/claude-code-subagents-collection
34. claude-code-agents (59 specialists)
59 specialized AI agents for development, including TypeScript, Python, Java, and Kotlin reviewers alongside workflow automation agents. Particularly well-suited for onboarding teams who need coverage across multiple languages. repo: https://github.com/baryhuang/claude-code-agents
35. multi-agent-squad
Production-ready multi-agent orchestration with task routing and consensus decision-making that works across coding agents. The architecture here is worth studying even if you are not ready to deploy it. repo: https://github.com/ruvnet/multi-agent-squad
36. Agent-Fusion
Multi-agent orchestration across AI tools: Claude Code, Codex CLI, Amazon Q, and Gemini, with bidirectional collaboration and intelligent task routing. If you are operating in a multi-tool environment, this is how you coordinate across them. repo: https://github.com/0xAsten/Agent-Fusion
37. sub-agents
A manager for Claude Code sub-agents that adds hooks and custom slash commands with a 175-star lightweight setup. The simplest entry point into subagent management without the complexity of a full orchestration system. repo: https://github.com/webdevtodayjason/sub-agents
38. sub-agents-directory
A curated directory of 100+ subagent prompts with MCP servers included and a community-maintained index. More discovery-focused than implementation-focused, making it a useful companion to the implementation repositories. repo: https://github.com/dotneet/sub-agents-directory
39. claude-code-heavy
Multi-agent research orchestration with parallel research agents built specifically for Claude Code. 72 stars. When you need to research something thoroughly and simultaneously rather than sequentially, this is the architecture. repo: https://github.com/iamcatdev/claude-code-heavy
40. ClaudeCodeAgents
A QA-focused agent collection with 455 stars, built around testing agents and validators. Drops into existing QA workflows and handles the part of development that most AI coding tools handle worst: ensuring that what was built actually works. repo: https://github.com/webdevtodayjason/ClaudeCodeAgents
## Category 5: MCP Servers
By default, Claude Code runs in your terminal. It reads files, executes commands, and interacts with APIs. That is the entire scope.
MCP servers extend that scope dramatically. They act as standardized bridges between Claude and the rest of your stack: your GitHub pull requests, your Postgres database, your browser, your Figma designs, your project management tools. Without MCP, Claude Code is a smart programmer's assistant. With MCP, it becomes a participant in your actual workflow.
41. claude-code-mcp
Run Claude Code itself as an MCP server, enabling the agent-in-agent pattern where Claude Code instances can recursively delegate coding tasks to other Claude Code instances. Integrates with Claude Desktop.
repo: https://github.com/steipete/claude-code-mcp
42. zilliztech/claude-context
Semantic code search MCP for massive codebases that pulls the right files without requiring manual breadcrumbing. Critical for anyone working in repositories large enough that context selection becomes a problem rather than an afterthought.
repo: https://github.com/zilliztech/claude-context
43. claude-context-mode
An MCP server that reduces context output by 98%, compressing 315 KB down to 5.4 KB per call. 2,200 stars. The token savings compound fast across a full working day, making this one of the highest-ROI repositories in the entire list. repo: https://github.com/d-e-s-o/claude-context-mode
44. Continuous-Claude-v2
Context management for Claude Code using ledger-based hooks that maintain state across sessions without polluting MCP execution context. 2,200 stars. Solves one of the fundamental limitations of session-based AI work. repo: https://github.com/continuous-claude/continuous-claude-v2
45. context7
Up-to-date, version-specific library and framework documentation injected directly into prompts in real time. Works with Cursor, VS Code, Claude Code, and Windsurf. If Claude is giving you outdated API suggestions, this is the fix. repo: https://github.com/upstash/context7
46. postgres-mcp
A safe PostgreSQL MCP server with read-only default mode, multi-connection support, and defense-in-depth security. The right way to give Claude access to your database without creating a security liability.
repo: https://github.com/crystaldba/postgres-mcp
47. playwright-mcp
Browser automation via MCP: navigate pages, fill forms, scrape dynamic content. Official Microsoft-backed. The bridge between Claude Code and any task that requires a real browser rather than an API.
repo: https://github.com/microsoft/playwright-mcp
48. mcp-server-sqlite
A lightweight SQLite MCP server for local development and prototyping. Simple install, official MCP team maintenance, and the fastest way to give Claude access to local data without setting up a full database.
repo: https://github.com/modelcontextprotocol/servers
49. claude-code-mcp-enhanced
An enhanced version of claude-code-mcp with more detailed instructions and emphatic prompts that improve task-following compliance. Built by grahama1970 for situations where the standard MCP server's direction needs reinforcement.
repo: https://github.com/grahama1970/claude-code-mcp-enhanced
50. danielrosehill/Claude-Code-MCP-List
A short, curated MCP collection focused specifically on Linux desktop Claude Code use cases. Battle-tested selections for practical daily work rather than theoretical completeness.
repo: https://github.com/danielrosehill/Claude-Code-MCP-List
## Category 6: Orchestration and Workflows
The next bottleneck after capability is you. You can only monitor one session at a time. You can only approve one plan. You can only be in one context.
Orchestration tools remove that bottleneck. They let you run parallel Claude sessions, coordinate multi-step workflows, preserve state between runs, and manage agents the way a team lead manages developers. This is where Claude Code stops being a productivity tool and starts being something closer to infrastructure.
51. Claude-Flow
An enterprise-grade AI orchestration platform with 11,400 stars and multi-agent orchestration with persistent memory. Claude-Flow represents a genuine architectural approach to building with AI at scale rather than just using it faster.
repo: https://github.com/ruvnet/claude-flow
52. claude-squad
A terminal multiplexer for AI coding agents with 5,600 stars. Manages Claude Code, Aider, Codex, OpenCode, and Amp simultaneously, allowing parallel agents without the chaos that parallel work normally creates.
repo: https://github.com/smtg-ai/claude-squad
53. claude_code_agent_farm
A parallel Claude Code sessions orchestrator designed for systematic codebase improvement at scale. Automation-first design for the kind of work that benefits from many simultaneous passes rather than one careful sequential one.
repo:https://github.com/Dicklesworthstone/claude_code_agent_farm
54. ccpm (Claude Code PM)
A project-management workflow for Claude Code by Ran Aroussi, combining specialized agents with slash commands in a turnkey setup. Bridges the gap between AI coding assistance and project management.
repo: https://github.com/automazeio/ccpm
55. cc-sessions
An opinionated extension set for Claude Code with 1,500 stars, covering hooks, subagents, commands, and task and git management infrastructure. Well-suited for teams who want a coherent workflow system rather than a collection of individual tools.
repo: https://github.com/cc-sessions/cc-sessions
56. claude-sessions
Custom slash commands for session tracking with 1,100 stars and comprehensive documentation. Focused specifically on dev session logging and the kind of context continuity that most Claude Code users manage manually.
repo: https://github.com/iannuttall/claude-sessions
57. claude-code-infrastructure-showcase
An infrastructure approach to skills with hooks that activate skills contextually based on what you are working on. Smart skill routing that adapts to any project without requiring explicit invocation.
repo: https://github.com/diet103/claude-code-infrastructure-showcase
58. infinite-agentic-loop
A demonstration of the infinite agentic loop pattern using a two-prompt system design. 492 stars. Experimental but genuinely educational for understanding the architectural ceiling of what Claude Code can become.
repo: https://github.com/disler/infinite-agentic-loop
59. agentsys
An automation harness for multiple tools with 473 stars, 14 plugins, 43 agents, and 30 skills spanning Claude Code, OpenCode, and Codex. One of the more comprehensive polyglot automation systems in the ecosystem.
repo: https://github.com/agentsys/agentsys
60. Citadel
An agent orchestration harness with 430 stars featuring parallel agent coordination and session memory persistence. Clean architecture that rewards teams who take the time to understand it before deploying it.
repo: https://github.com/gabrieltonnellier/citadel
## Category 7: Memory and Context
Every session starts from scratch. That is Claude Code's most significant limitation in practice and the one that creates the most daily friction.
The decisions you made last week. The architecture discussion you had yesterday. The reason you chose PostgreSQL over MongoDB. The bug pattern that appeared three times before you solved it. All of it is gone when you open a new terminal.
Memory systems solve this. They give Claude a persistent state across sessions so that today's Claude benefits from everything last week's Claude learned.
61. thedotmack/claude-mem
Long-term memory via compression that persists context between sessions. Lightweight, composable, and designed to solve the single most common complaint about Claude Code without adding complexity that gets in the way.
repo: https://github.com/thedotmack/claude-mem
62. hanfang/claude-memory-skill
A minimal hierarchical memory system using background agents and filesystem persistence. Low-friction setup that works alongside existing skills rather than replacing them. repo: https://github.com/hanfang/claude-memory-skill
63. 0xfurai/claude-subconscious
A background agent that watches your sessions, reads your files, and builds memory over time with 2,400 stars. The most elegant approach to memory on this list: you do not manage it, it manages itself.
repo: https://github.com/0xfurai/claude-subconscious
64. claude-code-semantic-memory
Persistent semantic memory for Claude Code using vector-based recall for session continuity. 70 stars. The technical approach here is different from filesystem-based memory and worth understanding if you are working on large codebases where semantic search matters. repo: https://github.com/dheerajoruganty/claude-code-semantic-memory
65. claude-user-memory
A user memory system with 12 specialist agents and intelligent orchestration that automatically selects the right specialist for each memory task. 125 stars. The most architecturally sophisticated memory system on the list.
repo: https://github.com/brennercruvinel/claude-user-memory
66. RepomixPack your entire codebase into one AI-readable file with 20,900 stars, outputting in XML, Markdown, or plain text. Essential for large-repo context. One of the most practically useful repositories on the entire list regardless of what else you install.
repo: https://github.com/yamadashy/repomix
67. graphifyAn interactive knowledge graph generated from your codebase that you can explore, search, and ask questions about. Works with Claude Code, Codex, and Cursor. Particularly powerful for onboarding into unfamiliar codebases.
repo: https://github.com/safishamsi/graphify
68. context-primeSystematic project context priming by elizaOS with scenario-specific priming commands and CLAUDE.md enhancement. Treats context injection as a systematic practice rather than an improvised workaround. repo: https://github.com/elizaOS/context-prime
69. claude-code-heavy-researchMulti-agent research with parallel context for deep-dive research workflows. When a research task is too large for a single agent pass, this is the architecture that handles it.
repo: https://github.com/iamcatdev/claude-code-heavy
70. ledgerA context-preserving ledger system using hooks to maintain ledger state and handle handoffs between agents without polluting the MCP execution context. The architecture behind Continuous-Claude-v2's session continuity.
repo: https://github.com/continuous-claude/continuous-claude-v2
## Category 8: Slash Commands and Hooks
Every time you type the same multi-sentence instruction for the third time, you are doing work that a slash command would have done for you. /fix-issue 456. /review. /tdd. One command instead of a paragraph, every single time.
Hooks go further. They fire automatically at key workflow moments: blocking commits that skip tests, formatting code before a push, injecting context at session start, enforcing the rules you would otherwise have to remember to state. You stop reminding Claude of your standards because the standards are enforced before you even start typing.
71. wshobson/commands
57 production-ready slash commands with 1,700 stars, organized across 15 workflows and 42 tools with multi-agent orchestration built in. The most comprehensive single slash command collection available.
repo: https://github.com/wshobson/commands
72. nizos/tdd-guard
Automated TDD enforcement for Claude Code with 1,700 stars. Blocks test-skipping and explains exactly why it blocked, making it both a quality gate and an educational tool for teams building the TDD habit.
repo: https://github.com/nizos/tdd-guard
73. ccundo
Granular undo for Claude Code with 1,300 stars. Provides action-level rollback with seamless integration, which changes the risk calculation for every experimental change you let Claude make.
repo: https://github.com/RonitSachdev/ccundo
74. claude-canvas
A TUI toolkit that gives Claude Code its own display with 1,100 stars. Provides an interactive terminal UI with rich visualization for situations where the standard terminal output is not enough.
repo: https://github.com/claude-canvas/claude-canvas
75. ClaudoPro Directory
A wide selection of hooks, commands, and subagents curated for specialized task coverage. Better quality control than average template sites. The ghost curation makes it a consistently reliable discovery source.
repo: https://github.com/ClaudoPro/claudopro-directory
76. danielrosehill/Claude-Slash-Commands
A curated collection of custom slash commands organized by category with a pre-push hook for sync and index-based discovery. Practical and well-maintained.
repo: https://github.com/danielrosehill/Claude-Slash-Commands
77. claude-commands/command-fix-issue
A TDD-based GitHub issue fixer via /fix-issue 456. Runs the full Red, Green, Refactor cycle and creates a pull request automatically. The workflow that most development teams should be using for issue resolution.
repo: https://github.com/claude-commands/command-fix-issue
78. Design Review Workflow
Automated UI/UX design review by Patrick Ellis covering responsive checks, accessibility audits, and specialized subagents for design-specific concerns. Fills the gap that most development-focused tooling leaves open. repo: https://github.com/patrick-ellis/design-review-workflow
79. tdd-implement
A TDD slash command by jerseycheese with Red/Green/Refactor automation, test-first enforcement, and feature-to-test breakdown. A focused implementation of the specific workflow that TDD guard protects.
repo: https://github.com/jerseycheese/tdd-implement
80. security-review
The official security review slash command that scans pull requests for issues and suggests fixes. Easy to disable when not needed. The simplest way to add automated security review to any workflow.
repo: https://github.com/anthropics/claude-code-security-review
## Category 9: Guides and Learning
Claude Code has real complexity that nobody warns you about. The command line is straightforward. But prompt engineering for code, skill creation, MCP configuration, hook design, context management, and token economics all have depth that takes time to develop.
Most developers get stuck, blame the tool, and move on. The repositories in this category exist to prevent that. Reference guides, pattern encyclopedias, system prompt archives, and training materials that would take months to develop independently are available here, now, for free.
81. ThamJiaHe/claude-prompt-engineering-guide
A definitive prompt engineering guide covering Opus 4.6, Sonnet 4.6, and Haiku 4.5 with 220+ verified sources. Covers Skills, Plugins, MCP, Hooks, and Ultrathink. The most comprehensive single reference document for serious Claude Code use.
repo: https://github.com/ThamJiaHe/claude-prompt-engineering-guide
82. nikiforovall/claude-code-handbook
Best practices for Claude Code workflows with distributable plugins included. Daily-use patterns and battle-tested tips that reflect real production experience rather than theoretical best practice. repo: https://github.com/nikiforovall/claude-code-handbook
83. FlorianBruniaux/claude-code-ultimate-guide
A massive 22,000+ line documentation project with 225 production templates, 9 AI personas, 26 slash commands, and 271 quiz questions. The most thorough single reference document in the Claude Code ecosystem.
repo: https://github.com/FlorianBruniaux/claude-code-ultimate-guide
84. Piebald-AI/claude-code-system-prompts
All Claude Code system prompts, updated within minutes of each new release. Includes tool descriptions, sub-agent prompts, and a changelog across 155+ versions. Essential for anyone who wants to understand how the tool actually works at the prompt level. repo: https://github.com/Piebald-AI/claude-code-system-prompts
85. Encyclopedia of Agentic Coding Patterns
Over 190 patterns for AI-assisted development by Wolf McNally, organized from foundational to agentic construction and covering governance and testing. The most systematic attempt to document what actually works in agentic coding.
repo: https://github.com/wolfmcnally/Encyclopedia-of-Agentic-Coding-Patterns
86. RiyaParikh0112/claude-code-playbook
A Claude Code playbook with TDD workflows, integration test patterns, CI/CD pipeline configurations, and PR preparation guides. Practical documentation oriented around shipping rather than learning. repo: https://github.com/RiyaParikh0112/claude-code-playbook
87. 40+ Claude Code Tips
A power-user tips collection covering custom status line scripts, system prompt halving tricks, and Claude Code in containers. The kind of practical knowledge that usually only exists in experienced developers' personal notes. repo: https://github.com/iannuttall/claude-code-tips
88. awesome-claude-prompts topic
Topic-based discovery of prompt repositories through community-submitted repos tagged for navigation. Useful for ongoing discovery as the ecosystem grows rather than one-time reference.
repo: https://github.com/topics/awesome-claude-prompts
89. Context Priming by disler
A systematic context priming approach with scenario-specific commands, project context patterns, and reference templates. Treats priming as a discipline rather than a habit.
repo: https://github.com/disler/context-priming
90. My Experience With Claude Code (2 weeks)
Real-world lessons and workflows from actual usage: TODO.md discipline, cost management, and refactor strategies. The authenticity of a practitioner's experience rather than the authority of official documentation.
repo: https://github.com/shaharia-lab/my-claude-code-journey
## Category 10: IDE Integrations and Desktop
Claude Code runs in the terminal. That is fine if you live in the terminal. It is less fine if you work primarily in an editor and want Claude integrated into your existing environment rather than running in a separate window that requires constant context-switching.
The repositories in this category bring Claude into the tools you already use: VS Code, Neovim, Emacs, and the desktop itself.
91. coder/claudecode.nvim
The first Neovim IDE integration for Claude Code, built in pure Lua with zero dependencies using a WebSocket-based MCP protocol reverse-engineered from the official extensions. If you live in Neovim, this is what makes Claude Code work there.
repo: https://github.com/coder/claudecode.nvim
92. greggh/claude-code.nvim
A seamless Neovim and Claude Code integration with a toggle terminal window, --continue and custom variants, and auto-reload of modified files. A more practical daily-use alternative to claudecode.nvim depending on your Neovim workflow.
repo: https://github.com/greggh/claude-code.nvim
93. manzaltu/claude-code-ide.el
An Emacs integration for Claude Code with ediff-based suggestions, LSP/flymake/flycheck diagnostics integration, and tree-sitter AST queries. The most technically sophisticated Emacs integration on the list.
repo: https://github.com/manzaltu/claude-code-ide.el
94. stevemolitor/claude-code.el
A simple Emacs interface for Claude Code CLI as a terminal wrapper with lightweight integration designed to be idiomatic Emacs rather than feature-heavy. The right choice for Emacs users who want minimal friction over maximum features.
repo: https://github.com/stevemolitor/claude-code.el
95. Haleclipse/Claudix
A VS Code extension for Claude Code with an interactive chat interface, session management, and a Vue 3 and TypeScript architecture. Brings Claude Code into the IDE that most developers spend the majority of their time in. repo: https://github.com/Haleclipse/Claudix
96. Clawd on Desk
A desktop pet that reacts to your Claude sessions in real time: thinking, typing, juggling, and sleeping based on what Claude Code is actually doing. Zero productivity value. Completely delightful. Included because the best ecosystems also have joy in them.
repo: https://github.com/Ruller-Lulu/clawd-on-desk
97. Claude Desktop Debian
An unofficial Claude desktop app for Linux as a Debian-native package, community-maintained to fill the gap that Anthropic has not officially addressed. Essential for Linux users who want a native desktop experience.
repo: https://github.com/aaddrick/claude-desktop-debian
98. instructkr/claw-code
A Korean-built Claude Code alternative and tooling collection with localized workflows and lightweight install. A community project that demonstrates the international reach of the ecosystem.
repo: https://github.com/instructkr/claw-code
## Category 11: Specialized Workflows
Some repositories resist categorization because they do one specific thing exceptionally well. A UI/UX skill that handles design patterns with a depth no general collection matches. A real-time directory that tracks the entire ecosystem as it evolves. Niche tools that solve exactly one problem better than anything else available.
These deserve attention precisely because they are specific.
99. nextlevelbuilder/ui-ux-pro-max-skill: A UI/UX skill for design-focused workflows with structured design patterns, component-oriented guidance, and production-grade output. The tool for developers who need Claude to think like a designer, not just write design-related code.
repo: https://github.com/nextlevelbuilder/ui-ux-pro-max-skill
100. ClaudeWaveA real-time directory of 3,800+ Claude repositories organized across 8 categories with tagged use-cases, ranked by stars and velocity, and updated every 12 hours. The living map of an ecosystem that is growing faster than any static list can track.
repo: https://github.com/claudewave/claudewave
## Where to Start
One hundred repositories is a lot to absorb. Most people who read a list this long install nothing.
Here is the honest path that actually produces results.
This week, install three things. Start with anthropics/skills to understand what a well-structured skill looks like at the official level. Then install Repomix because the ability to pack your codebase into one readable file is immediately useful regardless of everything else. Then install obra/superpowers because the brainstorm-to-merge workflow covers 80% of what you actually do in a development session.
Those three installations will show you, through direct experience, what is actually different about using Claude Code as a system rather than a tool. From there the next three choices become obvious based on what you actually find yourself wanting that you do not yet have.
The memory category matters most for developers working on long-running projects. The MCP servers category matters most for developers who need Claude to integrate with existing tooling. The orchestration category matters most for teams who want to run parallel workflows. The guides category matters most for developers who want to understand what is possible before they start building.
The ecosystem is evolving too quickly for anyone to have the complete picture. But the developers who have installed 10 to 20 of these repositories are already operating at a level that most Claude Code users, paying the same subscription, have never experienced and do not know is possible.
That gap is the opportunity. And it is available right now, today, for the cost of reading a list and running a few installs.
The tool you have been using is the default version.
These repositories are how you build the version that compounds.
## 相关链接
- [NeilXbt](https://x.com/neil_xbt)
- [@neil_xbt](https://x.com/neil_xbt)
- [6.9K](https://x.com/neil_xbt/status/2056386359376396458/analytics)
- [https://github.com/hesreallyhim/awesome-claude-code](https://github.com/hesreallyhim/awesome-claude-code)
- [https://github.com/ComposioHQ/awesome-claude-skills](https://github.com/ComposioHQ/awesome-claude-skills)
- [https://github.com/ComposioHQ/awesome-claude-plugins](https://github.com/ComposioHQ/awesome-claude-plugins)
- [https://github.com/langgptai/awesome-claude-prompts](https://github.com/langgptai/awesome-claude-prompts)
- [https://github.com/alvinunreal/awesome-claude](https://github.com/alvinunreal/awesome-claude)
- [https://github.com/VoltAgent/awesome-agent-skills](https://github.com/VoltAgent/awesome-agent-skills)
- [https://github.com/VoltAgent/awesome-claude-code-subagents](https://github.com/VoltAgent/awesome-claude-code-subagents)
- [https://github.com/ccplugins/awesome-claude-code-plugins](https://github.com/ccplugins/awesome-claude-code-plugins)
- [https://github.com/jqueryscript/awesome-claude-code](https://github.com/jqueryscript/awesome-claude-code)
- [https://github.com/danielrosehill/Claude-Code-Repos-Index](https://github.com/danielrosehill/Claude-Code-Repos-Index)
- [https://github.com/anthropics/claude-code](https://github.com/anthropics/claude-code)
- [https://github.com/anthropics/skills](https://github.com/anthropics/skills)
- [https://github.com/anthropics/claude-plugins](https://github.com/anthropics/claude-plugins)
- [https://github.com/anthropics/claude-code-sdk-python](https://github.com/anthropics/claude-code-sdk-python)
- [https://github.com/anthropics/claude-agent-sdk-typescript](https://github.com/anthropics/claude-agent-sdk-typescript)
- [https://github.com/anthropics/claude-code-security-review](https://github.com/anthropics/claude-code-security-review)
- [@claude](https://x.com/@claude)
- [https://github.com/anthropics/claude-code-action](https://github.com/anthropics/claude-code-action)
- [https://github.com/github/github-mcp-server](https://github.com/github/github-mcp-server)
- [https://github.com/anthropics/anthropic-cookbook](https://github.com/anthropics/anthropic-cookbook)
- [https://github.com/anthropics/courses](https://github.com/anthropics/courses)
- [https://github.com/obra/superpowers](https://github.com/obra/superpowers)
- [https://github.com/obra/superpowers-marketplace](https://github.com/obra/superpowers-marketplace)
- [https://github.com/obra/superpowers-skills](https://github.com/obra/superpowers-skills)
- [https://github.com/obra/superpowers-lab](https://github.com/obra/superpowers-lab)
- [https://github.com/affaan-m/everything-claude-code](https://github.com/affaan-m/everything-claude-code)
- [https://github.com/travisvn/awesome-claude-skills](https://github.com/travisvn/awesome-claude-skills)
- [https://github.com/trailofbits/claude-code-skills](https://github.com/trailofbits/claude-code-skills)
- [https://github.com/K-Dense-AI/claude-scientific-skills](https://github.com/K-Dense-AI/claude-scientific-skills)
- [https://github.com/ay-automate/ay-skills](https://github.com/ay-automate/ay-skills)
- [https://github.com/BehiSecc/awesome-claude-skills](https://github.com/BehiSecc/awesome-claude-skills)
- [https://github.com/wshobson/agents](https://github.com/wshobson/agents)
- [https://github.com/vijaythecoder/awesome-claude-agents](https://github.com/vijaythecoder/awesome-claude-agents)
- [https://github.com/davepoon/claude-code-subagents-collection](https://github.com/davepoon/claude-code-subagents-collection)
- [https://github.com/baryhuang/claude-code-agents](https://github.com/baryhuang/claude-code-agents)
- [https://github.com/ruvnet/multi-agent-squad](https://github.com/ruvnet/multi-agent-squad)
- [https://github.com/0xAsten/Agent-Fusion](https://github.com/0xAsten/Agent-Fusion)
- [https://github.com/webdevtodayjason/sub-agents](https://github.com/webdevtodayjason/sub-agents)
- [https://github.com/dotneet/sub-agents-directory](https://github.com/dotneet/sub-agents-directory)
- [https://github.com/iamcatdev/claude-code-heavy](https://github.com/iamcatdev/claude-code-heavy)
- [https://github.com/webdevtodayjason/ClaudeCodeAgents](https://github.com/webdevtodayjason/ClaudeCodeAgents)
- [https://github.com/steipete/claude-code-mcp](https://github.com/steipete/claude-code-mcp)
- [https://github.com/zilliztech/claude-context](https://github.com/zilliztech/claude-context)
- [https://github.com/d-e-s-o/claude-context-mode](https://github.com/d-e-s-o/claude-context-mode)
- [https://github.com/continuous-claude/continuous-claude-v2](https://github.com/continuous-claude/continuous-claude-v2)
- [https://github.com/upstash/context7](https://github.com/upstash/context7)
- [https://github.com/crystaldba/postgres-mcp](https://github.com/crystaldba/postgres-mcp)
- [https://github.com/microsoft/playwright-mcp](https://github.com/microsoft/playwright-mcp)
- [https://github.com/modelcontextprotocol/servers](https://github.com/modelcontextprotocol/servers)
- [https://github.com/grahama1970/claude-code-mcp-enhanced](https://github.com/grahama1970/claude-code-mcp-enhanced)
- [https://github.com/danielrosehill/Claude-Code-MCP-List](https://github.com/danielrosehill/Claude-Code-MCP-List)
- [https://github.com/ruvnet/claude-flow](https://github.com/ruvnet/claude-flow)
- [https://github.com/smtg-ai/claude-squad](https://github.com/smtg-ai/claude-squad)
- [https://github.com/Dicklesworthstone/claude_code_agent_farm](https://github.com/Dicklesworthstone/claude_code_agent_farm)
- [https://github.com/automazeio/ccpm](https://github.com/automazeio/ccpm)
- [https://github.com/cc-sessions/cc-sessions](https://github.com/cc-sessions/cc-sessions)
- [https://github.com/iannuttall/claude-sessions](https://github.com/iannuttall/claude-sessions)
- [https://github.com/diet103/claude-code-infrastructure-showcase](https://github.com/diet103/claude-code-infrastructure-showcase)
- [https://github.com/disler/infinite-agentic-loop](https://github.com/disler/infinite-agentic-loop)
- [https://github.com/agentsys/agentsys](https://github.com/agentsys/agentsys)
- [https://github.com/gabrieltonnellier/citadel](https://github.com/gabrieltonnellier/citadel)
- [https://github.com/thedotmack/claude-mem](https://github.com/thedotmack/claude-mem)
- [https://github.com/hanfang/claude-memory-skill](https://github.com/hanfang/claude-memory-skill)
- [https://github.com/0xfurai/claude-subconscious](https://github.com/0xfurai/claude-subconscious)
- [https://github.com/dheerajoruganty/claude-code-semantic-memory](https://github.com/dheerajoruganty/claude-code-semantic-memory)
- [https://github.com/brennercruvinel/claude-user-memory](https://github.com/brennercruvinel/claude-user-memory)
- [https://github.com/yamadashy/repomix](https://github.com/yamadashy/repomix)
- [https://github.com/safishamsi/graphify](https://github.com/safishamsi/graphify)
- [https://github.com/elizaOS/context-prime](https://github.com/elizaOS/context-prime)
- [https://github.com/iamcatdev/claude-code-heavy](https://github.com/iamcatdev/claude-code-heavy)
- [https://github.com/continuous-claude/continuous-claude-v2](https://github.com/continuous-claude/continuous-claude-v2)
- [https://github.com/wshobson/commands](https://github.com/wshobson/commands)
- [https://github.com/nizos/tdd-guard](https://github.com/nizos/tdd-guard)
- [https://github.com/RonitSachdev/ccundo](https://github.com/RonitSachdev/ccundo)
- [https://github.com/claude-canvas/claude-canvas](https://github.com/claude-canvas/claude-canvas)
- [https://github.com/ClaudoPro/claudopro-directory](https://github.com/ClaudoPro/claudopro-directory)
- [https://github.com/danielrosehill/Claude-Slash-Commands](https://github.com/danielrosehill/Claude-Slash-Commands)
- [https://github.com/claude-commands/command-fix-issue](https://github.com/claude-commands/command-fix-issue)
- [https://github.com/patrick-ellis/design-review-workflow](https://github.com/patrick-ellis/design-review-workflow)
- [https://github.com/jerseycheese/tdd-implement](https://github.com/jerseycheese/tdd-implement)
- [https://github.com/anthropics/claude-code-security-review](https://github.com/anthropics/claude-code-security-review)
- [https://github.com/ThamJiaHe/claude-prompt-engineering-guide](https://github.com/ThamJiaHe/claude-prompt-engineering-guide)
- [https://github.com/nikiforovall/claude-code-handbook](https://github.com/nikiforovall/claude-code-handbook)
- [https://github.com/FlorianBruniaux/claude-code-ultimate-guide](https://github.com/FlorianBruniaux/claude-code-ultimate-guide)
- [https://github.com/Piebald-AI/claude-code-system-prompts](https://github.com/Piebald-AI/claude-code-system-prompts)
- [https://github.com/wolfmcnally/Encyclopedia-of-Agentic-Coding-Patterns](https://github.com/wolfmcnally/Encyclopedia-of-Agentic-Coding-Patterns)
- [https://github.com/RiyaParikh0112/claude-code-playbook](https://github.com/RiyaParikh0112/claude-code-playbook)
- [https://github.com/iannuttall/claude-code-tips](https://github.com/iannuttall/claude-code-tips)
- [https://github.com/topics/awesome-claude-prompts](https://github.com/topics/awesome-claude-prompts)
- [https://github.com/disler/context-priming](https://github.com/disler/context-priming)
- [https://github.com/shaharia-lab/my-claude-code-journey](https://github.com/shaharia-lab/my-claude-code-journey)
- [https://github.com/coder/claudecode.nvim](https://github.com/coder/claudecode.nvim)
- [https://github.com/greggh/claude-code.nvim](https://github.com/greggh/claude-code.nvim)
- [https://github.com/manzaltu/claude-code-ide.el](https://github.com/manzaltu/claude-code-ide.el)
- [https://github.com/stevemolitor/claude-code.el](https://github.com/stevemolitor/claude-code.el)
- [https://github.com/Haleclipse/Claudix](https://github.com/Haleclipse/Claudix)
- [https://github.com/Ruller-Lulu/clawd-on-desk](https://github.com/Ruller-Lulu/clawd-on-desk)
- [https://github.com/aaddrick/claude-desktop-debian](https://github.com/aaddrick/claude-desktop-debian)
- [https://github.com/instructkr/claw-code](https://github.com/instructkr/claw-code)
- [https://github.com/nextlevelbuilder/ui-ux-pro-max-skill](https://github.com/nextlevelbuilder/ui-ux-pro-max-skill)
- [https://github.com/claudewave/claudewave](https://github.com/claudewave/claudewave)
- [Upgrade to Premium](https://x.com/i/premium_sign_up)
- [10:48 PM · May 18, 2026](https://x.com/neil_xbt/status/2056386359376396458)
- [6,998 Views](https://x.com/neil_xbt/status/2056386359376396458/analytics)
- [View quotes](https://x.com/neil_xbt/status/2056386359376396458/quotes)
---
*导出时间: 2026/5/19 09:18:33*
---
## 中文翻译
# 如果你还没用这 100 个仓库，那你就是在错误地使用 Claude Code。
**作者**: NeilXbt
**日期**: 2026-05-18T14:48:08.000Z
**来源**: [https://x.com/neil_xbt/status/2056386359376396458](https://x.com/neil_xbt/status/2056386359376396458)
---
大多数人只把 Claude Code 当作一个智能终端助手。
你输入指令。它写代码。你审查。继续。确实有用。但这和你以前的做法没有本质区别，只是快了一点。
我以前也是这么想的。直到我发现，真正的 Claude Code——那个让其他开发者用来构建整个公司的版本，那个能运行 30 个并行代理、在会话间保持记忆、自动强制执行 TDD、集成你技术栈中所有工具、并且能力每周都在复合增长的版本——并不是你开箱即用的那个版本。
它是通过组装仓库得到的版本。
Claude Code 周边的生态系统发展之快，几乎没人能跟得上。新的技能、插件、子代理、MCP 服务器、编排器、记忆系统、IDE 集成和工作流工具每天都在涌现。它们中的大多数从未上过 Hacker News。大多数都没得到它们应得的 Star。而大多数 Claude Code 用户从未听说过其中的任何一个。
这种差距——即 Claude Code 默认的样子与你在其基础上构建之后的样子之间的差距——正是产生生产力差异的地方。那些使用同样的订阅却效率高出 5 倍的开发者，并不是更聪明。也不是工作更努力。他们只是安装了几个仓库，并构建了一个系统，而不是仅仅在使用一个工具。
本文涵盖了 11 个类别中的 100 个最佳仓库。每一个都是真实、维护良好且可立即使用的。读完本文，你将确切知道该安装什么以及为什么要安装。
## 类别 1：精选列表与元索引
在你安装任何具体工具之前，你需要一张地图。Claude Code 生态系统的演进速度太快，单个人无法追踪。精选列表通过众包筛选来解决这一问题，依据实际质量而非 GitHub 星标进行过滤，并紧跟这个每周都在变化的生态系统。
从这里开始。经常回来看看。
1. **awesome-claude-code**
整个 Claude Code 生态系统的权威精选列表。这是拥有 28,500 星的主索引，拥有卓越的质量控制机制：只允许 Claude 本身提交拉取请求。技能、钩子、命令、编排器和插件都在这里，经过筛选以确保真正有用，而非仅仅存在。仓库：https://github.com/hesreallyhim/awesome-claude-code
2. **awesome-claude-skills (ComposioHQ)**
超过 50 个用于自动化的生产级 Claude 技能，涵盖 PostgreSQL、深度研究、CSV 处理和根因追溯。与 Claude Code、Codex、Cursor 和 Gemini CLI 跨兼容。通过 /skill add 直接安装。仓库：https://github.com/ComposioHQ/awesome-claude-skills
3. **awesome-claude-plugins (ComposioHQ)**
精选的插件集合，涵盖命令、代理、钩子和 MCP 服务器。包括 connect-apps、frontend-design 和 artifacts-builder。通过 /plugin marketplace add 安装。仓库：https://github.com/ComposioHQ/awesome-claude-plugins
4. **awesome-claude-prompts**
拥有 4,200 星的优化 Claude 提示词集合，使用 XML 结构化提示技术。涵盖业务、创意和开发工作流，第一天即可复制粘贴使用。仓库：https://github.com/langgptai/awesome-claude-prompts
5. **awesome-claude**
关于 Anthropic Claude 相关内容的全方位列表：课程、指南、桌面应用、文章、MCP 覆盖、BMAD 方法和 Claude Code 技巧。对于想了解全貌后再深入细节的新手来说，这是最好的起点。仓库：https://github.com/alvinunreal/awesome-claude
6. **awesome-agent-skills (VoltAgent)**
超过 1,000 个源自真实工程团队而非 AI 生成的代理技能。久经沙场考验，兼容 Claude Code、Codex、Gemini CLI 和 Cursor。包括 Notion、WhatsApp、NotebookLM 和记忆技能。仓库：https://github.com/VoltAgent/awesome-agent-skills
7. **awesome-claude-code-subagents (VoltAgent)**
针对每种技术栈和任务的手工精选子代理，组织成特定领域的专业知识包。生产就绪的组件，可与大多数编码代理集成。仓库：https://github.com/VoltAgent/awesome-claude-code-subagents
8. **awesome-claude-code-plugins (ccplugins)**
专注于 Claude Code 插件：斜杠命令、子代理、MCP 服务器和钩子，每个条目都包含安装命令。同时也支持托管你自己的市场。仓库：https://github.com/ccplugins/awesome-claude-code-plugins
9. **awesome-claude-code (jqueryscript)**
精选的工具、IDE 集成和框架集合，带有星标标记。涵盖 SDK、安全工具和代理集合。作为发现生态系统中不那么显眼部分的工具非常出色。仓库：https://github.com/jqueryscript/awesome-claude-code
10. **Claude-Code-Repos-Index**
75+ 个优质 Claude Code 仓库的索引目录，全部原创，无 AI 生成的填充内容，由 Daniel Rosehill 维护。生态系统的一个实用导航器，专注于日常可用性而非新奇事物。仓库：https://github.com/danielrosehill/Claude-Code-Repos-Index
## 类别 2：Anthropic 官方仓库
在看任何社区仓库之前，先从这里开始。这些是 Anthropic 的官方项目：CLI 本身、SDK、技能库和 GitHub Action。此列表中的每个社区仓库都构建在它们之上。
它们可能并不总是拥有最广泛的功能范围，但对于需要稳定性的团队来说，它们是最稳定、文档最完善、最可靠的选择。
11. **anthropics/claude-code**
官方的 Claude Code 终端工具，拥有 55,000 星。针对代码库的自然语言命令，由 Anthropic 直接维护。如果你在使用 Claude Code，你已经在使用它了。但该仓库本身包含文档、问题追踪和官方指导，而大多数用户从未阅读过。仓库：https://github.com/anthropics/claude-code
12. **anthropics/skills**
官方的 Agent Skills 公开仓库，拥有 37,500 星。涵盖 PDF、DOCX、XLSX、PPTX 和艺术生成。这是权威的技能参考，也是开始理解结构良好的技能实际样子的起点，然后你再开始编写自己的技能。仓库：https://github.com/anthropics/skills
13. **anthropics/claude-plugins**
由 Anthropic 管理的优质插件目录，拥有 2,800 星。一个经过审查的集合，对于那些需要知道安装的内容已经过工具构建者审查的团队来说，这提供了一个可靠的标准。仓库：https://github.com/anthropics/claude-plugins
14. **anthropics/claude-code-sdk-python**
官方的 Claude Code Python SDK，拥有 6,100 星。提供对功能的编程访问，并支持 Agent SDK 模式。对于任何在 Claude Code 之上构建自动化或集成的人来说，这都是必不可少的。仓库：https://github.com/anthropics/claude-code-sdk-python
15. **anthropics/claude-agent-sdk-typescript**
Claude Agent 的官方 TypeScript SDK，具有类型安全的工具编排、流式支持、MCP 集成，以及具有 send() 和 stream() 模式的 V2 预览版。任何在 Claude 上构建生产级 TypeScript 工具的团队的基础。仓库：https://github.com/anthropics/claude-agent-sdk-typescript
16. **anthropics/claude-code-security-review**
AI 驱动的安全审查 GitHub Action，拥有 2,800 星。分析拉取请求的差异以查找漏洞，并无缝集成到现有的 CI/CD 管道中。这是整个列表中最实用的仓库之一。仓库：https://github.com/anthropics/claude-code-security-review
17. **anthropics/claude-code-action**
用于拉取请求中 Claude 的官方 GitHub Action。@claude 提及会触发分析、差异审查和补丁建议，所有这些都遵循你自己仓库的约定。仓库：https://github.com/anthropics/claude-code-action
18. **github/github-mcp-server**
官方的 GitHub MCP 服务器。通过 X-MCP-Readonly 以只读模式直接访问仓库、拉取请求、问题和 CI/CD。整个 MCP 生态系统中使用最广泛的服务器。仓库：https://github.com/github/github-mcp-server
19. **anthropic-cookbook**
官方的 Anthropic 食谱，包含有效的配方：笔记本、代理模式、工具使用示例、RAG、分类和摘要。由 Anthropic 团队维护，是了解什么是良好的 API 使用的最值得信赖的来源。仓库：https://github.com/anthropics/anthropic-cookbook
20. **courses (Anthropic)**
免费的官方 Anthropic 课程，涵盖提示工程基础、API 使用、代理和 AI 流利度。直接来自源头，也是在你开始尝试社区工具之前建立正确心理模型的最快方式。仓库：https://github.com/anthropics/courses
## 类别 3：技能集合
技能是无需编写一行代码即可扩展 Claude Code 功能的最快方式。它们是 Markdown 文件，用于教导 Claude 如何执行特定任务：编写拉取请求、在 Remotion 中创建视频、进行安全审计、生成图表、运行 SEO 分析。
一个好的技能的力量在于，它将 Claude 从通用助手转变为按需提供的领域专家。你不必每次会话都从头开始解释你的技术栈。技能已经知道了。
21. **obra/superpowers**
最受欢迎的 Claude Code 技能库，拥有 148,000 星。涵盖完整的开发周期：头脑风暴、规范、计划、TDD、审查和合并，内置了子代理编排。如果你只安装一个技能库，那就是它了。仓库：https://github.com/obra/superpowers
22. **obra/superpowers-marketplace**
Superpowers 生态系统的官方市场，拥有超过 20 个久经沙场考验的技能。包括 /brainstorm、/write-plan 和 /execute-plan，以及用于无缝工作流连续性的 SessionStart 上下文注入。仓库：https://github.com/obra/superpowers-marketplace
23. **obra/superpowers-skills**
Superpowers 的社区可编辑技能库，自动克隆到 ~/.config/superpowers/skills/。Fork 它，提交 PR，并为整个 Superpowers 社区使用的技能做出贡献。生态系统的协作骨干。仓库：https://github.com/obra/superpowers-skills
24. **obra/superpowers-lab**
正在积极开发中的实验性技能，包括语义重复函数检测和尖端技术。这是 Superpowers 下一版本的诞生地。即使你还没准备好使用实验性功能，也值得关注。仓库：https://github.com/obra/superpowers-lab
25. **everything-claude-code**
Cerebral Valley x Anthropic 黑客马拉松获奖作品，基于 10 多个月的实际使用构建。包含 30 个代理、136 个技能、60 个斜杠命令和 1,282 个测试，覆盖率达 98%。这是列表中最全面的单一仓库之一。仓库：https://github.com/affaan-m/everything-claude-code
26. **travisvn/awesome-claude-skills**
社区最佳技能集合，拥有 22,000+ 次安装，涵盖 SEO、营销、设计、安全和研究。包括针对不受信任技能的安全警告，这正是让社区集合值得信赖的筛选纪律所在。
仓库：https://github.com/travisvn/awesome-claude-skills
27. **trailofbits/claude-code-skills**
来自 Trail of Bits 的安全研究和审计技能，这是业内最受尊敬的安全工程公司之一。如果你正在做任何与漏洞检测相关的工作，这个集合正是为此而构建的。
仓库：https://github.com/trailofbits/claude-code-skills
28. **K-Dense-AI/claude-scientific-skills**
专为科学计算设计的技能，拥有用于研究工作流、学术任务和数据分析的专业库。大多数技能集合缺失的部分是科学研究。这填补了这一空白。仓库：https://github.com/K-Dense-AI/claude-scientific-skills
29. **ay-automate/ay-skills**
十个经过生产测试的技能，涵盖 SEO 审计、无头浏览器自动化、Remotion 视频生成和图表。开箱即用，专注于大多数技能集合未涉及的任务。仓库：https://github.com/ay-automate/ay-skills
30. **BehiSecc/awesome-claude-skills**
一个作为列表之列表的元资源，涵盖知识获取、更深层次的思维框架和技能自动化实用程序。最适合那些已经探索过主要集合并想进一步深入的人。仓库：https://github.com/BehiSecc/awesome-claude-skills
## 类别 4：代理和子代理
单一的 Claude 会话有限制。随着上下文的增长，性能会下降。最终你会得到一个对太多事情一知半解的单个代理，而不是一组各有所长的专家。
子代理在结构上解决了这个问题。你不再是一个 Claude 吃力地处理二十个任务，而是拥有一个团队：每个代理都有自己的上下文窗口、自己的领域专业知识和单一的明确职责。编排器负责路由任务。专家负责执行任务。
31. **wshobson/agents**
拥有 25,000 星的庞大生产级子代理集合，跨越策略、开发、安全、设计、数据和研究领域。可与 Claude Code 工作流直接兼容。这是列表中 Star 最多、使用最广泛的仓库之一。仓库：https://github.com/wshobson/agents
32. **awesome-claude-agents**
一组专门设计用于协作完成完整功能的专家级 AI 代理团队。3,700 星。这些代理具有跨不同技术栈的专家级知识，旨在协同工作而非孤立存在。仓库：https://github.com/vijaythecoder/awesome-claude-agents
33. **davepoon/claude-code-subagents-collection**
超过 100 个生产就绪的子代理，拥有 2,200 星和全面的领域专业知识。专为与任何 Claude Code 工作流进行直接集成而构建。仓库：https://github.com/davepoon/claude-code-subagents-collection
34. **claude-code-agents (59 specialists)**
59 个用于开发的专业 AI 代理，包括 TypeScript、Python、Java 和 Kotlin 审查员，以及工作流自动化代理。特别适合需要覆盖多种语言的入职团队。仓库：https://github.com/baryhuang/claude-code-agents
35. **multi-agent-squad**
具有任务路由和共识决策的生产级多代理编排，可跨编码代理工作。这里的架构即使你还没准备好部署，也值得研究。仓库：https://github.com/ruvnet/multi-agent-squad
36. **Agent-Fusion**
跨 AI 工具的多代理编排：Claude Code、Codex CLI、Amazon Q 和 Gemini，具有双向协作和智能任务路由。如果你在多工具环境中操作，这就是你协调它们的方式。仓库：https://github.com/0xAsten/Agent-Fusion
37. **sub-agents**
Claude Code 子代理的管理器，添加了钩子和自定义斜杠命令，设置轻量级，拥有 175 星。这是进入子代理管理最简单的入口点，没有完整编排系统的复杂性。仓库：https://github.com/webdevtodayjason/sub-agents
38. **sub-agents-directory**
包含 100+ 子代理提示词的精选目录，包含 MCP 服务器和社区维护的索引。与专注于实现相比，它更侧重于发现，因此它是实现仓库的有用伴侣。仓库：https://github.com/dotneet/sub-agents-directory
39. **claude-code-heavy**
专为 Claude Code 构建的多代理研究编排，具有并行研究代理。72 星。当你需要彻底且同时地研究某事而不是