GitHub Trending Daily — 2026-05-18: 에이전트 스택이 운영 소프트웨어와 붙었다
Daily snapshot of GitHub Trending, captured at KST 2026-05-18.
오늘의 GitHub Trending은 단순히 “에이전트가 뜬 날”이 아니다. openhuman이 여전히 1위를 유지하는 가운데, CLI-Anything, agent-skills, scientific-agent-skills, codegraph, cal.diy가 함께 올라오면서 개인용 AI 런타임 + 에이전트용 스킬 + 로컬 지식 + 운영 소프트웨어가 하나의 묶음으로 보인다.
Snapshot
| ★ today | Repo | Total ★ | Signal |
|---|---|---|---|
| +1,694 | tinyhumansai/openhuman | 13,036 | Personal AI runtime keeps the consumer AI OS angle alive |
| +923 | tech-leads-club/agent-skills | 3,488 | Skill registry, not just a skill bundle |
| +908 | oven-sh/bun | 91,701 | Runtime/tooling still matters outside the AI niche |
| +860 | colbymchenry/codegraph | 3,253 | Local code knowledge graph for Claude Code |
| +764 | K-Dense-AI/scientific-agent-skills | 23,792 | Domain-specific skills are now ordinary infrastructure |
| +704 | Anil-matcha/Open-Generative-AI | 15,058 | Self-hosted image/video generation widens the chart |
| +425 | calcom/cal.diy | 43,238 | Scheduling infrastructure for everyone |
| +405 | BigBodyCobain/Shadowbroker | 7,038 | Agent-facing intelligence aggregation |
Trend signal
1) openhuman stays the anchor
openhuman is still the clearest “personal AI runtime” signal on the board. The point is not the model itself. The point is that the user-facing layer is becoming a persistent operational surface with memory, tools, and a UI that can absorb more than a single chat session.
2) Skills are moving from artifact to registry
CLI-Anything, agent-skills, and scientific-agent-skills do not tell the same story, but they all point in one direction: skills are no longer just prompt snippets. They are being packaged, validated, installed, and reused like software.
3) Trending is no longer only AI
cal.diy, Ghost, listmonk, plausible, and medusa show a second current inside the same chart. The market still rewards software that handles scheduling, publishing, newsletters, analytics, and commerce. In practice, agent tools need to live inside that stack, not beside it.
4) Local knowledge and local control are growing together
codegraph, DreamServer, Shannon, and Shadowbroker all sit closer to private workflows than public demos. They suggest a stronger bias toward local-first knowledge graphs, local AI, and actionable intelligence. That matters because agent usefulness rises sharply when it can stay close to code and private data.
Why this matters
The trend line is now less about “which model won” and more about which operational layer owns the workflow.
| Layer | Examples | What it means |
|---|---|---|
| Runtime | openhuman, bun | A stable execution surface matters as much as model quality |
| Skills | agent-skills, scientific-agent-skills, CLI-Anything | Reusable procedures are becoming deployable assets |
| Local knowledge | codegraph, DreamServer | Private context and local search are entering the default stack |
| Operations | cal.diy, Ghost, listmonk, plausible, medusa | Real software products still dominate attention outside pure AI |
This is the practical signal worth tracking: agent tooling is converging with software that already runs businesses.
오늘의 상위 프로젝트를 짧게 읽기
tinyhumansai/openhuman— 개인 AI 런타임 신호가 계속 가장 강하다.HKUDS/CLI-Anything— 소프트웨어를 agent-native로 바꾸는 방향이 꾸준히 먹힌다.tech-leads-club/agent-skills— 검증된 스킬 레지스트리라는 표현이 중요하다. 단순 모음집이 아니다.K-Dense-AI/scientific-agent-skills— 도메인별 스킬은 이제 실험이 아니라 인프라다.colbymchenry/codegraph— Claude Code 주변에서 로컬 지식 그래프 수요가 분명히 생겼다.calcom/cal.diy— 에이전트가 붙기 쉬운 운영 소프트웨어가 여전히 강하다.
마치며
어제까지의 트렌드는 “에이전트 하네스”와 “메모리”에 더 가까웠다면, 오늘은 그 위에 운영 소프트웨어가 덧붙었다. 이건 좋은 신호다. 실제로 오래 쓰이는 AI 제품은 모델보다도 런타임, 스킬, 로컬 지식, 운영 계층을 얼마나 잘 묶느냐에 따라 결정되기 때문이다.
openhuman 같은 개인용 런타임, agent-skills 같은 재사용 가능한 작업 체계, codegraph 같은 로컬 지식 구조가 동시에 뜨는 건 우연이 아니다. GitHub Trending은 지금 AI 기능보다 AI를 놓고 돌아가는 소프트웨어 구조를 더 크게 밀어주고 있다.