Refly is an open-source platform for building and deploying agent skills, positioning itself as an alternative to traditional workflow automation tools and code-first agent frameworks. Written in TypeScript, the repository implements what the project describes as the first open-source agent skills builder, enabling users to define skills through vibe workflow and execute them across multiple platforms including Claude Code, Cursor, and various bot frameworks.
The core philosophy of Refly centers on treating skills as durable infrastructure rather than prompts. The platform uses a Model-Native DSL that allows users to describe business logic in natural language, which the system then compiles into deterministic, reusable, and versioned agent capabilities. This approach aims to bridge the gap between raw APIs and intelligent agents, enabling organizations to codify messy business logic into structured, version-controlled skills that can be invoked with high reliability.
Refly's primary use cases span multiple deployment scenarios. Users can build APIs for Lovable, create webhooks for Lark and Feishu integration, export skills to Claude Code, and construct Clawdbot instances. The platform emphasizes rapid deployment, claiming users can transition from a static enterprise standard operating procedure to a production-ready agent skill in under three minutes. The execution model includes an intervenable runtime that allows users to pause, audit, and redirect agent logic during execution, addressing concerns about black-box AI behavior and ensuring operational compliance.
The repository integrates with an extensive ecosystem of tools and services. The platform supports over 3,000 native tools including Stripe, Slack, Salesforce, and GitHub, while maintaining full compatibility with Model Context Protocol servers. Users can also connect private skill connectors to databases, scripts, and internal systems. The skill registry serves as a central repository for managing, versioning, and sharing agent capabilities across teams.
According to GitGenius activity data, the repository shows strong community engagement with 202 tracked issues and pull requests. The median response latency for issues and PRs is 0.0 hours, indicating rapid triage, while the mean response time is 118.8 hours. The most active labels are enhancement with 82 items, good first issue with 58 items, and bug with 32 items. Primary contributors include mrcfps with 334 events, pftom with 111 events, and CH1111 with 40 events. The repository shares overlapping contributors with langgenius/dify, apitable/apitable, and lobehub/lobehub, suggesting connections within the broader AI and automation ecosystem.
The platform positions itself distinctly against existing solutions by offering capabilities that legacy automation tools like n8n and Dify lack, such as intervenable runtime and copilot-led construction, while providing more accessible deployment than code-first frameworks like LangChain. Refly emphasizes production-ready stateful APIs, hot-fix capabilities during execution, and portability across different agent environments. The project targets both individual builders seeking rapid prototyping and enterprises requiring scalable skills governance and reliable agent infrastructure deployment across their organizations.
