Top Agent Skill Repositories
Across all agent runtimes, obra/superpowers leads on overall score, followed by anthropics/skills and garrytan/gstack. K-Dense-AI/scientific-agent-skills is the strongest domain-specific collection. All entries verified live and scored against the public rubric; metrics fetched 2026-07-07.
The ranking
| # | Entry | Overall | Tier | Reliability | Leverage | Last commit |
|---|---|---|---|---|---|---|
| 1 | obra/superpowers | 83 | A-Tier | 84 | 90 | 2026-07-06 |
| 2 | promptfoo/promptfoo | 83 | A-Tier | 78 | 82 | 2026-07-07 |
| 3 | langchain-ai/langgraph | 82 | A-Tier | 80 | 82 | 2026-07-06 |
| 4 | pydantic/pydantic-ai | 82 | A-Tier | 82 | 78 | 2026-07-07 |
| 5 | anthropics/skills | 81 | A-Tier | 78 | 82 | 2026-07-01 |
| 6 | garrytan/gstack | 80 | A-Tier | 80 | 86 | 2026-06-25 |
| 7 | confident-ai/deepeval | 80 | A-Tier | 76 | 80 | 2026-07-06 |
| 8 | langfuse/langfuse | 80 | A-Tier | 76 | 78 | 2026-07-07 |
| 9 | microsoft/playwright-mcp | 80 | A-Tier | 80 | 84 | 2026-06-29 |
| 10 | modelcontextprotocol/servers | 79 | A-Tier | 76 | 84 | 2026-07-06 |
| 11 | upstash/context7 | 79 | A-Tier | 76 | 80 | 2026-07-07 |
| 12 | openai/openai-agents-python | 79 | A-Tier | 78 | 76 | 2026-07-07 |
| 13 | github/github-mcp-server | 79 | A-Tier | 78 | 80 | 2026-07-07 |
| 14 | mem0ai/mem0 | 78 | A-Tier | 72 | 80 | 2026-07-07 |
| 15 | google/adk-python | 78 | A-Tier | 74 | 76 | 2026-07-07 |
| 16 | K-Dense-AI/scientific-agent-skills | 77 | A-Tier | 70 | 84 | 2026-07-06 |
| 17 | letta-ai/letta | 77 | A-Tier | 74 | 78 | 2026-07-03 |
| 18 | oraios/serena | 77 | A-Tier | 74 | 82 | 2026-07-06 |
| 19 | guardrails-ai/guardrails | 77 | A-Tier | 74 | 76 | 2026-06-30 |
| 20 | crewAIInc/crewAI | 76 | A-Tier | 72 | 78 | 2026-07-07 |
| 21 | microsoft/autogen | 75 | A-Tier | 74 | 78 | 2026-04-15 |
| 22 | davila7/claude-code-templates | 75 | A-Tier | 70 | 78 | 2026-07-07 |
| 23 | getzep/zep | 74 | A-Tier | 72 | 74 | 2026-07-06 |
| 24 | topoteretes/cognee | 74 | A-Tier | 70 | 76 | 2026-07-07 |
| 25 | wshobson/agents | 72 | A-Tier | 68 | 76 | 2026-07-06 |
| 26 | basicmachines-co/basic-memory | 71 | A-Tier | 68 | 72 | 2026-07-07 |
| 27 | eyaltoledano/claude-task-master | 71 | A-Tier | 70 | 78 | 2026-04-28 |
| 28 | github/awesome-copilot | 71 | A-Tier | 64 | 72 | 2026-07-07 |
| 29 | danielmiessler/Fabric | 68 | B-Tier | 62 | 74 | 2026-06-09 |
| 30 | openai/evals | 68 | B-Tier | 66 | 68 | 2026-04-14 |
| 31 | protectai/llm-guard | 68 | B-Tier | 68 | 70 | 2025-12-15 |
| 32 | SuperClaude-Org/SuperClaude_Framework | 68 | B-Tier | 66 | 74 | 2026-06-13 |
| 33 | gohypergiant/agent-skills | 66 | Experimental | 66 | 62 | 2026-07-06 |
| 34 | VoltAgent/awesome-claude-code-subagents | 64 | B-Tier | 58 | 68 | 2026-06-24 |
| 35 | Pimzino/claude-code-spec-workflow | 63 | B-Tier | 64 | 72 | 2025-09-07 |
| 36 | snarktank/ai-dev-tasks | 63 | B-Tier | 62 | 70 | 2025-11-05 |
| 37 | multica-ai/andrej-karpathy-skills | 61 | Experimental | 60 | 70 | 2026-04-20 |
| 38 | hesreallyhim/awesome-claude-code | 59 | B-Tier | 46 | 62 | 2026-07-07 |
| 39 | punkpeye/awesome-mcp-servers | 58 | B-Tier | 44 | 60 | 2026-07-04 |
| 40 | qdhenry/Claude-Command-Suite | 58 | B-Tier | 60 | 64 | 2026-03-01 |
| 41 | PatrickJS/awesome-cursorrules | 57 | B-Tier | 45 | 60 | 2026-05-30 |
| 42 | VoltAgent/awesome-agent-skills | 56 | B-Tier | 42 | 58 | 2026-06-30 |
| 43 | sanjeed5/awesome-cursor-rules-mdc | 55 | B-Tier | 48 | 60 | 2026-05-19 |
All published repository entries, ranked by overall score. Partially-scored entries are listed but flagged; experimental-tier entries (under 6 months old) are scored but not tiered against established ones. Full rubric.
Top entries in detail
A process-first skill framework for Claude Code: brainstorming, test-driven development, systematic debugging, and verification skills that…
Test harness for prompts, agents, and RAG pipelines: declarative test cases, assertions, matrix comparisons, and red-team scanning, runnable…
The most adopted graph-based agent orchestration framework: workflows as explicit state machines with checkpointing, human-in-the-loop…
Frequently asked questions
How often does this ranking update?
Automated metrics refresh nightly; editorial scores are re-reviewed at least every 90 days for S and A tiers, and immediately on major releases or credible reports.
Can a maintainer pay to move up?
No. No score, tier, rank, or editorial verdict has ever been or will ever be for sale. Sponsorship, when it exists, is a separate labeled surface.