Last Updated: July 10, 2026 | 29,000+ repositories surveyed | 50 curated picks across 6 categories
- Overview
- Quick Navigation
- Top Repositories by Category
- Comprehensive Statistics
- Use Case Mapping
- Getting Started
- Contributing
Six categories that define the current AI agent ecosystem:
| Category | What it covers |
|---|---|
| Agent Skills | Frameworks, libraries, and collections of reusable agent capabilities |
| Artifact Registries | Platforms for discovering, storing, and managing agent artifacts |
| A2A Multi-Agent Workflows | Orchestration frameworks for agent-to-agent communication and coordination |
| Swarm Agents | Multi-agent systems designed for collaborative, distributed execution |
| Agent Governance | Policy enforcement, compliance, and control frameworks for autonomous agents |
| TokenOps | Token optimization, cost tracking, and LLM budget management |
⚠️ Data note: Several entries in the source list show "TBD" for stars/forks — these appear to be newer or unindexed repositories. Treat those figures as unverified; confirm directly on GitHub before relying on them.
| Category | Repos | Jump to |
|---|---|---|
| Agent Skills | 8 | Section 1 |
| Artifact Registries | 8 | Section 2 |
| A2A Multi-Agent Workflows | 8 | Section 3 |
| Swarm Agents | 10 | Section 4 |
| Agent Governance | 10 | Section 5 |
| TokenOps | 6 | Section 6 |
| Language | Count |
|---|---|
| Python | 28 |
| TypeScript | 13 |
| Go | 3 |
| JavaScript | 2 |
| Multi-language | 2 |
| Rust | 1 |
| Shell | 1 |
| Tier | Projects |
|---|---|
| 100,000+ Stars | Superpowers, Anthropic Skills |
| 25,000–100,000 Stars | Agent-Skills (Addy Osmani), Agentic Awesome Skills, Scientific Agent Skills, OpenAI Agents Python, Awesome Agent Skills, Vercel Labs Skills |
| 10,000–25,000 Stars | Google Skills, Nanobrowser, Microsoft Agent Framework |
Modular, reusable capabilities that extend AI agents' functionality.
1. Anthropic Skills — Public Repository for Agent Skills
Python • ⭐ 159,988 • 🍴 18,883
Official Anthropic repository for agent skills, containing production-grade skill implementations and serving as the foundational reference for Claude Code and other Anthropic-powered agents.
Best for: Teams wanting the canonical, officially maintained skill definitions Key features: Official Anthropic maintenance · production-grade implementations · reference standard for skill design
2. Agent-Skills by Addy Osmani — Production Engineering Skills
JavaScript • ⭐ 76,403 • 🍴 8,202
Production-grade engineering skills for AI coding agents, widely adopted across Claude Code, Cursor, and Codex ecosystems.
Best for: Software engineering teams standardizing agent behavior across tools Key features: Cross-tool compatibility · real-world engineering focus · large community adoption
3. Vercel Labs Skills — The Open Agent Skills Tool
TypeScript • ⭐ 25,702 • 🍴 2,152
Open-source agent skills tool with an npx skills command-line interface and tight integration into web development workflows.
Best for: Web/JS developers wanting a zero-friction CLI for skill installation
Key features: npx skills CLI · open framework · web-dev-native integration
4. Superpowers — Agentic Skills Framework
Shell • ⭐ 251,371 • 🍴 22,437
The largest repository in this registry — an agentic skills framework and software development methodology with massive community experimentation.
Best for: Exploring cutting-edge, community-driven agent capability design Key features: Highest star count in registry · large fork/experimentation base · methodology-level framework
5. Agentic Awesome Skills — Library of 1,900+ Skills
Python • ⭐ 42,766 • 🍴 6,796
Installable GitHub library of 1,900+ agentic skills compatible with Claude Code, Cursor, and Codex CLI, including an installer CLI and both official and community bundles.
Best for: Teams wanting the broadest possible skill library in one install Key features: 1,900+ skills · installer CLI · official + community bundles
6. Scientific Agent Skills — AI Scientist Capabilities
Python • ⭐ 30,618 • 🍴 3,081
Transforms AI agents into AI Scientists with 140 ready-to-use skills and 100+ scientific databases spanning biology, chemistry, medicine, and drug discovery.
Best for: Research teams building domain-specific scientific agents Key features: 140 scientific skills · 100+ database integrations · adoption reported across 160,000+ scientists
7. Awesome Agent Skills — Curated Collection
Multi-language • ⭐ 27,740 • 🍴 2,970
Curated collection of 1,000+ agent skills from official dev teams and the community, compatible with Claude Code, Codex, Gemini CLI, and Cursor.
Best for: Discovering skills across the widest range of agent platforms Key features: 1,000+ curated skills · multi-platform compatibility · active community curation
8. Google Skills — Google Products Integration
Python • ⭐ 14,494 • 🍴 1,112
Official Google agent skills for Google products and technologies, integrating deeply with Workspace, Cloud, and AI services.
Best for: Teams building agents inside the Google ecosystem Key features: Official Google maintenance · Workspace/Cloud integration · enterprise-grade reliability
Centralized platforms for discovering, versioning, and distributing agent artifacts.
9. ARA Registry Spec — Package Registry Specification
Rust • ⭐ TBD • 🍴 TBD
Package registry specification for AI development artifacts — agents, MCP servers, prompts, powers, and skills — establishing unified distribution standards.
Best for: Standards-setters building interoperable artifact ecosystems Key features: Formal spec (not just an implementation) · covers agents, MCP servers, prompts & skills
10. Agent Artifact Registry — Artifact Lineage System
Python • ⭐ TBD • 🍴 TBD
Artifact lineage and integrity subsystem preserving reusable project knowledge, memory candidates, and execution evidence.
Best for: Enterprises needing provenance tracking and reproducibility Key features: Lineage tracking · execution evidence capture · reproducibility focus
11. Agentic Registry — Gateway-Neutral Registry
Go • ⭐ TBD • 🍴 TBD
Gateway-neutral registry for agentic artifacts — skills, tools, MCPs, prompts, workflows, and blueprints — built for cross-platform interoperability.
Best for: Avoiding vendor lock-in across agent frameworks Key features: Gateway-neutral design · broad artifact type support · cross-platform portability
12. AgentX Platform — MCP Servers & Skills Registry
TypeScript • ⭐ TBD • 🍴 TBD
Open-source AI agent artifact registry for MCP servers, skills, and commands with version control and dependency resolution.
Best for: Teams managing MCP server sprawl Key features: Version control · discovery & dependency resolution · TypeScript backend
13. AgentHub — Central ADP-Compatible Registry
Multi-language • ⭐ TBD • 🍴 TBD
Central registry for ADP-compatible agents and ADPKG artifacts, enabling versioning, discovery, distribution, provenance, and governance across teams.
Best for: Cross-team enterprise agent artifact management Key features: ADP/ADPKG compatibility · provenance tracking · governance built in
14. Nika Registry — Content-Pinned Artifacts
Python • ⭐ TBD • 🍴 TBD
Shares workflows, packs, skills, and agents with content-pinning and CI verification — every entry re-proven by a conformance oracle.
Best for: Teams that want trust to live in the artifact, not the gatekeeper Key features: Content-pinning · CI-verified conformance oracle · novel trust model
15. LoopSkill API — Self-Hostable Registry
Python • ⭐ TBD • 🍴 TBD
Open-core "Spotify for AI agents" — a self-hostable registry of runnable agent artifacts (skills, bundles, loops, personalities), deployable via Docker Compose in 60 seconds.
Best for: Teams wanting a self-hosted, MCP-native artifact catalog Key features: 60-second Docker Compose setup · MCP-native · zero signup barrier · MPL-2.0 licensed
16. Hermes SQLite Toolkit — Artifact Cache System
Python • ⭐ TBD • 🍴 TBD
Zero-dependency SQLite toolkit providing tool cache, artifact registry, decision log, and experience tracking for local artifact persistence.
Best for: Lightweight, local-first artifact and memory persistence Key features: Zero dependencies · SQLite-based · decision logging · powers Hermes Agent's memory system
Orchestration frameworks enabling agent-to-agent communication and complex workflows.
17. OpenAI Agents Python — Lightweight Framework
Python • ⭐ 27,789 • 🍴 4,290
Official OpenAI framework for multi-agent workflows — a lightweight, intuitive Python API that has become a reference implementation for agent orchestration patterns.
Best for: Teams standardizing on OpenAI's official orchestration approach Key features: Official OpenAI maintenance · lightweight API · reference-grade orchestration patterns
18. Microsoft Agent Framework — Multi-Language Support
Python • ⭐ 12,006 • 🍴 2,016
Framework for building, orchestrating, and deploying AI agents with multi-agent workflows, supporting both Python and .NET.
Best for: Enterprises standardized on Microsoft's stack Key features: Python + .NET support · enterprise adoption at Microsoft · cross-platform coordination
19. Mission Control — Self-Hosted Orchestration
TypeScript • ⭐ 5,676 • 🍴 958
Self-hosted AI agent orchestration platform for dispatching tasks, running multi-agent workflows, monitoring spend, and governance from one dashboard.
Best for: Teams wanting a unified, self-hosted ops dashboard Key features: Self-hosted · spend monitoring · governance built in · unified dashboard
20. LLMStack — No-Code Multi-Agent Framework
Python • ⭐ 2,304 • 🍴 348
No-code framework to build LLM agents, workflows, and applications with custom data, via a visual workflow builder.
Best for: Non-engineers building agent workflows visually Key features: No-code visual builder · custom data support · accessible agent composition
21. Solace Agent Mesh — Event-Driven Framework
Python • ⭐ 4,963 • 🍴 268
Event-driven framework for building and orchestrating multi-agent AI systems, integrating seamlessly with real-world data sources.
Best for: Event-driven architectures needing agent coordination Key features: Event-driven design · real-world data source integration · complex multi-step workflows
22. Nanobrowser — Chrome Extension Automation
TypeScript • ⭐ 13,444 • 🍴 1,411
Open-source Chrome extension for AI-powered web automation, running multi-agent workflows with your own LLM API key — positioned as an OpenAI Operator alternative.
Best for: Browser-based automation without a hosted-agent subscription Key features: Bring-your-own API key · browser-native agents · strong community adoption
23. Forge — Self-Hosted LLM Workflows
Python • ⭐ 2,164 • 🍴 163
Python framework for self-hosted LLM tool-calling and multi-step agentic workflows, built for local-first orchestration.
Best for: Local-first, self-hosted agent workflow deployments Key features: Self-hosted · tool-calling support · multi-step workflow patterns
24. OpenAgentsControl — Plan-First Development
TypeScript • ⭐ 4,527 • 🍴 362
AI agent framework for plan-first development workflows with approval-based execution, supporting TypeScript, Python, Go, and Rust.
Best for: Teams wanting review/approval gates before agent execution Key features: Plan-first, approval-gated execution · multi-language support (TS, Python, Go, Rust)
Multi-agent systems designed for collaborative and distributed execution at scale.
25. OpenAI Swarm — Educational Framework
Python • ⭐ TBD • 🍴 TBD
Educational framework exploring ergonomic, lightweight multi-agent orchestration, managed by OpenAI's Solution team as a reference implementation.
Best for: Learning swarm orchestration fundamentals Key features: Official OpenAI reference · lightweight & ergonomic · simplicity-focused
26. Agency Swarm — Reliable Multi-Agent Framework
Python • ⭐ TBD • 🍴 TBD
Reliable multi-agent orchestration framework with production-ready error handling and resilience patterns.
Best for: Production swarms that need built-in resilience Key features: Error handling · resilience patterns · widely adopted for complex coordination
27. Swarms — Enterprise-Grade Framework
Python • ⭐ TBD • 🍴 TBD
Enterprise-grade, production-ready multi-agent orchestration framework supporting 23+ LLM providers.
Best for: Enterprises needing broad LLM-provider flexibility at scale Key features: 23+ LLM providers · enterprise-grade tooling · dedicated site (swarms.ai)
28. ElizaOS — Open Agentic OS
TypeScript • ⭐ TBD • 🍴 TBD
Open-source agentic operating system providing foundational infrastructure for agent swarms, with persistent memory and reasoning.
Best for: Building a full OS-level foundation for autonomous agents Key features: OS-level infrastructure · persistent memory · growing ecosystem
29. ClawTeam — Agent Swarm Intelligence
Python • ⭐ TBD • 🍴 TBD
"One Command → Full Automation" — hierarchical agent coordination with specialized agent roles for complex multi-agent tasks.
Best for: Single-command automation of multi-role agent tasks Key features: Hierarchical coordination · specialized agent roles · one-command automation
30. SwarmClaw — Self-Hosted Runtime
TypeScript • ⭐ TBD • 🍴 TBD
Open-source, self-hosted AI agent runtime and multi-agent framework with memory, MCP tools, schedules, delegation, and 23+ LLM providers.
Best for: Teams wanting a self-hosted Claude Code/LangChain alternative Key features: Self-hosted · MCP tools · scheduling & delegation · 23+ LLM providers
31. SwarmZero — Agent SDK
Python • ⭐ TBD • 🍴 TBD
SDK for building AI agents and swarms of agents, production-ready with debugging, monitoring, and enterprise support.
Best for: Developers who want an SDK-first swarm-building experience Key features: Production-ready SDK · built-in debugging/monitoring · enterprise support
32. Pentest Swarm AI — Security Agent Swarms
Go • ⭐ TBD • 🍴 TBD
Autonomous penetration testing via a swarm of AI agents, orchestrating recon, classification, exploitation, and reporting specialists with ReAct reasoning.
Best for: Security teams automating bug bounty, monitoring, or CTF workflows Key features: ReAct-based reasoning · 7+ native security tools · bug bounty & CTF modes
33. SwarmVault — Local-First Knowledge Graph
TypeScript • ⭐ TBD • 🍴 TBD
Local-first LLM wiki and open-source knowledge graph builder serving as an agent memory store — an Obsidian alternative for AI second-brain use cases.
Best for: Durable, shared memory across swarm operations Key features: Local-first · knowledge graph builder · RAG-ready agent memory
34. LangGraph Swarm — LangChain Integration
Python • ⭐ TBD • 🍴 TBD
LangGraph's swarm implementation enabling sophisticated agent coordination patterns, tightly integrated with the LangChain ecosystem.
Best for: Teams already invested in LangChain/LangGraph Key features: Native LangChain integration · production-ready orchestration · sophisticated coordination patterns
Policy enforcement, compliance, and control frameworks for autonomous agents.
35. Agent Governance Toolkit — Microsoft Framework
Python • ⭐ 4,788 • 🍴 839
Policy enforcement, zero-trust identity, execution sandboxing, and reliability engineering — covers all 10 of the OWASP Agentic Top 10.
Best for: Enterprises needing OWASP-aligned agent security governance Key features: Full OWASP Agentic Top 10 coverage · zero-trust identity · execution sandboxing
36. DefenseClaw — Security Governance
Python • ⭐ 766 • 🍴 136
Cisco's security governance framework for agentic AI, defending against 10+ OWASP Agentic vulnerabilities.
Best for: Security-conscious organizations wanting a vendor-backed defense layer Key features: Cisco-backed · 10+ OWASP vulnerability coverage · agent-specific threat defense
37. YAO Meta Skill — Engineering & Evaluation
Python • ⭐ 2,060 • 🍴 191
"Yielding AI Outcomes" — rigorous engineering, evaluation, governance, and portability system for reusable agent skills.
Best for: Managing the full lifecycle and compliance of agent skills Key features: Structured skill lifecycle governance · evaluation rigor · portability focus
38. Lunar Gateway — MCP Security Gateway
TypeScript • ⭐ 468 • 🍴 38
Agent-native MCP gateway acting as a proxy between agents and external systems, centralizing governance of tool access.
Best for: Centralizing control over what tools agents can reach Key features: MCP-native proxy · centralized tool-access governance · security-first design
39. EventCatalog — Domain Governance
TypeScript • ⭐ 2,759 • 🍴 267
Discovery and governance layer for event-driven systems, documenting domains, services, events, and schemas for both teams and AI agents.
Best for: Making event-driven architecture legible to AI agents Key features: Domain/service/schema documentation · agent-aware design · discovery layer
40. DashClaw — Governance Runtime
TypeScript • ⭐ 284 • 🍴 50
Governance runtime that intercepts agent actions, enforces guard policies, requires approvals, and produces audit-ready decision trails.
Best for: Compliance-driven organizations needing audit trails Key features: Action interception · approval requirements · audit-ready decision trails
41. Node9 Proxy — Execution Security Layer
TypeScript • ⭐ 208 • 🍴 17
Execution security layer providing deterministic "Sudo" governance and audit logs for autonomous AI agents.
Best for: Enforcing deterministic permission boundaries on agent actions Key features: Deterministic "Sudo" model · audit logging · execution-layer security
42. Evolving Agents — Autonomous Ecosystems
Python • ⭐ 452 • 🍴 35
Toolkit for autonomous, evolving agent ecosystems — create, execute, govern, and evolve agents that learn and operate within guardrails.
Best for: Governing agents that adapt or learn over time Key features: Governance for adaptive/learning agents · guardrail enforcement · ecosystem-level tooling
43. Traccia — OpenTelemetry Observability
Python • ⭐ 80 • 🍴 19
OpenTelemetry-native observability, governance, and compliance instrumentation for AI agents and LLM applications.
Best for: Teams standardizing on OpenTelemetry for agent observability Key features: OpenTelemetry-native · structured instrumentation · compliance-oriented monitoring
44. Harness Books — Design Philosophies
Python • ⭐ 2,616 • 🍴 308
Two books on harness engineering — the design philosophies behind Claude Code and Codex, covering constraints, query loops, context governance, and multi-agent verification.
Best for: Learning governance design patterns, not just using a tool Key features: Educational (not a library) · covers context governance & verification patterns · design-philosophy depth
Token optimization, cost tracking, and LLM budget management for AI agents.
45. Nasiko — Developer Control Plane with TokenOps
Python • ⭐ 3,605 • 🍴 546
Developer control plane for AI agents with built-in TokenOps support — token tracking and cost optimization alongside an MCP gateway.
Best for: Teams wanting agent management and cost control in one platform Key features: Combined control plane + TokenOps · MCP gateway integration · tokenomics tracking
46. TokenOps by KlarLabs — Rate-Limit Prediction
Go • ⭐ TBD • 🍴 TBD
Predicts rate-limit cutoffs inside your AI agent via a local MCP server and CLI, tracking Claude Max, ChatGPT plans, GitHub Copilot, and Cursor plan windows.
Best for: Avoiding surprise rate-limit cutoffs across multiple subscriptions Key features: Multi-provider rate-limit prediction · local MCP server + CLI · Docker-ready
47. TokenOps Cost Lab — Cost Engineering
JavaScript • ⭐ TBD • 🍴 TBD
Browser-local AI usage economics lab for API pricing, GitHub Copilot AI Credits, and agentic workload forecasting — 100% client-side.
Best for: Modeling cost scenarios without sending data to a backend Key features: 100% client-side · verified pricing-change tracking · workload forecasting
48. TokenOps Guardian — Budget Enforcement
Python • ⭐ 1 • 🍴 TBD
Autonomous agent budget-enforcement gatekeeper for enterprise LLM ingestion streams, mitigating runaway token costs at the runtime layer.
Best for: Hard-stopping runaway token spend in enterprise pipelines Key features: Runtime-layer enforcement · runaway-cost mitigation · enterprise ingestion focus
49. TokenOps Smart Proxy — Model Routing
Python • ⭐ TBD • 🍴 TBD
Autonomous cost-intelligence agent built with FastAPI that routes requests to the cheapest viable model, enforcing budget/rate policies and caching responses.
Best for: Automatically routing to the cheapest model that meets the bar Key features: Smart model routing · response caching · real-time cost-analytics API
50. Nerve — Adaptive Inference Control Plane
TypeScript • ⭐ TBD • 🍴 TBD
Adaptive inference control plane optimizing token usage through intelligent model selection, inference caching, and load balancing.
Best for: Optimizing inference cost/latency trade-offs automatically Key features: Adaptive model selection · inference caching · integrates tokenops, verifier & benchmarking
Figures below are recomputed directly from the 50 entries above. Repos marked "TBD" in the source data are excluded from sums and averages (noted per row).
| Metric | Value |
|---|---|
| Total repositories surveyed (source claim) | 29,000+ |
| Repositories in this document | 50 |
| Repos with reported star counts | 28 of 50 |
| Total combined stars (known repos only) | ~720,000 |
| Total combined forks (known repos only) | ~77,900 |
| Average stars per repo (known repos only) | ~25,700 |
| Category | Repos | Repos w/ known stars | Avg stars (known) |
|---|---|---|---|
| Agent Skills | 8 | 8 | ~78,600 |
| Artifact Registries | 8 | 0 | N/A — all TBD |
| A2A Multi-Agent Workflows | 8 | 8 | ~9,100 |
| Swarm Agents | 10 | 0 | N/A — all TBD |
| Agent Governance | 10 | 10 | ~1,450 |
| TokenOps | 6 | 2 | ~1,800 |
| Language | Repos |
|---|---|
| Python | 28 |
| TypeScript | 13 |
| Go | 3 |
| JavaScript | 2 |
| Multi-language | 2 |
| Rust | 1 |
| Shell | 1 |
| Rank | Repository | Stars | Category |
|---|---|---|---|
| 1 | Superpowers | 251,371 | Agent Skills |
| 2 | Anthropic Skills | 159,988 | Agent Skills |
| 3 | Agent-Skills (Addy Osmani) | 76,403 | Agent Skills |
| 4 | Agentic Awesome Skills | 42,766 | Agent Skills |
| 5 | Scientific Agent Skills | 30,618 | Agent Skills |
| 6 | OpenAI Agents Python | 27,789 | A2A Workflows |
| 7 | Awesome Agent Skills | 27,740 | Agent Skills |
| 8 | Vercel Labs Skills | 25,702 | Agent Skills |
| 9 | Google Skills | 14,494 | Agent Skills |
| 10 | Nanobrowser | 13,444 | A2A Workflows |
| If you need to... | Look at |
|---|---|
| Extend agent capability with reusable, packaged skills | Anthropic Skills, Agentic Awesome Skills, Awesome Agent Skills |
| Discover, version, and distribute agent artifacts | ARA Registry Spec, AgentHub, LoopSkill API |
| Coordinate agents talking to each other in a workflow | OpenAI Agents Python, Microsoft Agent Framework, Mission Control |
| Run large collaborative swarms at scale | Swarms, SwarmZero, Agency Swarm |
| Enforce policy, approvals, and audit trails on autonomous agents | Agent Governance Toolkit, DefenseClaw, DashClaw |
| Control or reduce LLM token spend | Nasiko, TokenOps Smart Proxy, TokenOps Guardian |
| Automate browser or security testing with agents | Nanobrowser, Pentest Swarm AI |
- Pick a category from the Quick Navigation or Use Case Mapping table that matches your problem.
- Compare 2–3 candidates — for repos with known stars, treat star/fork counts as rough adoption signals, not quality guarantees.
- Verify "TBD" entries directly on GitHub before adopting — 22 of the 50 repos here have unconfirmed metrics in the source data.
- Check for official backing (Anthropic, OpenAI, Microsoft, Google, Cisco) if you need long-term maintenance guarantees.
- Prototype before committing — most non-"Agent Skills" categories here are early-stage, single-maintainer projects.
- Agent Skills: github.com/search?q=agent+skills&type=repositories
- Artifact Registries: github.com/search?q=artifact+registry+agent&type=repositories
- Multi-Agent Workflows: github.com/search?q=multi+agent+workflow&type=repositories
- Swarm Agents: github.com/search?q=swarm+agents&type=repositories
- Agent Governance: github.com/search?q=agent+governance&type=repositories
- TokenOps: github.com/search?q=tokenops+agent&type=repositories
To suggest additions or improvements to this registry: report new repositories or categories, suggest description updates, propose new analysis angles, or flag emerging trends via issue or pull request.
- Data source: GitHub Search API, agent-related keyword searches
- Snapshot date: July 10, 2026 — stars/forks drift over time and "TBD" entries were unverified at compile time
- Compilation method: Top results by relevance/stars per category
- License: Provided as-is for research purposes; each repository retains its own license