The open-source AI agent framework ecosystem is experiencing explosive growth, with two major projects gaining significant traction. The agent harness system ECC has amassed over 304,000 stars on GitHub—making it one of the most starred AI projects—and added 718 stars today alone. Simultaneously, Google Labs launched Stitch Skills, a library of agent capabilities designed around the emerging Agent Skills open standard, which has already attracted nearly 10,000 stars. Both projects focus on enabling coding agents like Claude Code, Cursor, and others to work with extensible skill sets.
The convergence around shared standards marks a significant shift toward interoperability in the AI agent ecosystem. Rather than each tool building isolated agent frameworks, projects are aligning on common APIs and skill definitions that allow different agents and models to leverage the same tooling. This standardization enables developers to write agent skills once and deploy them across multiple platforms—whether users are running Claude Code, Google's agents, or other implementations. For the open-source community, this represents infrastructure maturation: moving from fragmented point solutions to a coherent, shared ecosystem.
This trend has practical implications for local and self-hosted deployments. By establishing open standards, developers can run agent-based tools without vendor lock-in, mix and match components, and maintain control over their infrastructure. The projects' focus on open-source licensing (MIT for ECC, Apache-2.0 for Stitch Skills) further enables community contribution and adaptation. As multi-agent systems become increasingly sophisticated—evidenced by complementary projects like PaperOrchestra for research automation—standardized skill frameworks become essential. This emerging ecosystem suggests a future where local, composable agent tooling becomes as accessible and extensible as traditional open-source software.
