Top GitHub repos for Claude Code (May 2026)
- Reference
- Tooling
25 hand-picked GitHub repos that drive the Claude Code ecosystem forward: harness, awesome lists, memory, token optimization, multi-agents, plugins. Data verified on GitHub 2026-05-11.
TL;DR
- The Claude Code ecosystem now exceeds 2 million cumulative stars across the 25 repos below (GitHub data as of 2026-05-11).
- 8 categories cover the real needs: harness, awesome lists, plugins, memory, token optimization, multi-agents, templates, and system prompt leaks.
- Three repos dominate their category:
anthropics/claude-code(122k★ official),everything-claude-code(179k★ harness leader),x1xhlol/system-prompts-and-models-of-ai-tools(137k★ internals reference). - A dozen more stand out for their clearly identified niche rather than star count:
claude-mem(memory),caveman(token reduction),graphify(knowledge graph),gstack(Garry Tan's templates). - List refreshed monthly. Dead or inactive repos (no commit in 90+ days) are removed.
Why this page exists
When an ecosystem explodes, you need a map. GitHub awesome lists (43k+ ★ each) are already great, but they list everything without prioritization. This page filters: we keep only repos that have both social traction (stars) and real-world usage value (clear description, recent commits, active community).
Each entry shows the GitHub star count measured on the day of publication. Numbers move fast (sometimes +10k★ per week on viral repos), so they're a rough order-of-magnitude signal, not absolute truth.
1. Harness & frameworks
"Harness" refers to the layer of prompts, skills, and hooks that turn Claude Code into a specialized assistant. The densest category.
| Repo | ⭐ | Why it matters |
|---|---|---|
anthropics/claude-code (ouvre un nouvel onglet) | 122,570 | The official Anthropic repo, source of truth for the CLI itself. GitHub issues + releases. |
affaan-m/everything-claude-code (ouvre un nouvel onglet) | 179,423 | The leading harness: skills, instincts, memory, security, research-first. The most complete one. |
forrestchang/andrej-karpathy-skills (ouvre un nouvel onglet) | 125,124 | A single CLAUDE.md derived from Karpathy's observations on LLM pitfalls. Required reading. |
garrytan/gstack (ouvre un nouvel onglet) | 93,773 | Garry Tan's exact setup (Y Combinator): 23 opinionated tools, defined roles (CEO, designer, QA…). |
shareAI-lab/learn-claude-code (ouvre un nouvel onglet) | 59,784 | "Bash is all you need" — a nano educational harness to understand how a CLI agent works. |
2. Awesome lists & curation
The reference annuaire pages to discover the ecosystem. Bookmark these.
| Repo | ⭐ | Why it matters |
|---|---|---|
punkpeye/awesome-mcp-servers (ouvre un nouvel onglet) | 86,707 | THE reference list for MCP servers. Sorted by category. |
ComposioHQ/awesome-claude-skills (ouvre un nouvel onglet) | 59,255 | Curation of public skills with usage examples. |
hesreallyhim/awesome-claude-code (ouvre un nouvel onglet) | 43,382 | Global inventory: skills, hooks, slash-commands, agents, plugins. |
sickn33/antigravity-awesome-skills (ouvre un nouvel onglet) | 37,180 | 1,400+ installable skills for Claude Code, Cursor, Codex CLI, Gemini. |
3. Plugins & marketplaces
The official Anthropic directory and community plugin ecosystems.
| Repo | ⭐ | Why it matters |
|---|---|---|
anthropics/claude-plugins-official (ouvre un nouvel onglet) | 19,104 | The official Anthropic-managed directory. Validated and maintained plugins. |
4. Memory & persistent context
To avoid losing context between sessions, one of Claude Code's main native gaps.
| Repo | ⭐ | Why it matters |
|---|---|---|
thedotmack/claude-mem (ouvre un nouvel onglet) | 74,819 | Captures everything the agent does, compresses, replays at startup. The dominant pattern. |
safishamsi/graphify (ouvre un nouvel onglet) | 46,618 | Turns a code folder into a queryable knowledge graph. Works with Cursor, Codex too. |
5. Token optimization & cost
When the Anthropic bill climbs. Concrete patterns to reduce token consumption.
| Repo | ⭐ | Why it matters |
|---|---|---|
JuliusBrussee/caveman (ouvre un nouvel onglet) | 58,237 | The "why use many token when few token do trick" skill. Cuts 65% tokens by talking caveman. |
musistudio/claude-code-router (ouvre un nouvel onglet) | 33,820 | Decides which model handles what based on context. Switches Haiku → Sonnet → Opus on the fly. |
6. Multi-agents & orchestration
When a single agent isn't enough. Swarm, teams, and multi-Claude orchestration patterns.
| Repo | ⭐ | Why it matters |
|---|---|---|
ruvnet/ruflo (ouvre un nouvel onglet) | 49,014 | Leading multi-agent orchestration platform. Smart swarms and autonomous workflows. |
wshobson/agents (ouvre un nouvel onglet) | 35,205 | Smart automation + multi-agent orchestration in Python. |
Yeachan-Heo/oh-my-claudecode (ouvre un nouvel onglet) | 33,417 | Teams-first multi-agent orchestration (defined roles, explicit communication). |
7. Templates & meta-prompting
Opinionated setups ready to clone.
| Repo | ⭐ | Why it matters |
|---|---|---|
gsd-build/get-shit-done (ouvre un nouvel onglet) | 61,537 | Meta-prompting + context engineering + spec-driven dev. Lightweight but powerful. |
shanraisshan/claude-code-best-practice (ouvre un nouvel onglet) | 52,365 | "From vibe coding to agentic engineering" — pattern accumulation. |
davila7/claude-code-templates (ouvre un nouvel onglet) | 27,168 | CLI to configure and monitor Claude Code. Fast bootstrap. |
luongnv89/claude-howto (ouvre un nouvel onglet) | 32,369 | Visual guide with step-by-step examples and copy-paste templates. |
8. System prompts & internals
To understand how Claude Code works under the hood. Essential for advanced prompting.
| Repo | ⭐ | Why it matters |
|---|---|---|
x1xhlol/system-prompts-and-models-of-ai-tools (ouvre un nouvel onglet) | 137,157 | THE reference leak: system prompts of Claude Code, Cursor, Devin, Junie, Lovable, etc. |
asgeirtj/system_prompts_leaks (ouvre un nouvel onglet) | 40,039 | Extracted system prompts: ChatGPT, Claude Opus 4.7/4.6/Sonnet, Claude Code, Gemini. |
How to use this
Three typical use cases:
- Deciding on a harness: start with
everything-claude-code(most complete) orgstack(defined roles, simpler). Don't stack two different harnesses. - Filling a specific gap: need memory?
claude-mem. Need to cut the bill?caveman+claude-code-router. Need to orchestrate multiple agents?ruflo. - Understanding internals: read
system-prompts-and-models-of-ai-toolsonce, then revisit the patterns described in the awesome lists.
Next steps
- Compare the best public skills:
/ecosystem/awesome-skills(coming, story ECO-3). - Discover the official Anthropic plugins:
/ecosystem/awesome-plugins(coming, ECO-4). - Explore MCP servers by category:
/ecosystem/awesome-mcp-servers(coming, ECO-5). - Learn how to create your own skill:
/skills/create-custom. - Dig into memory and persistent context:
/prompting/claude-md.