Claude Leak Confirms It: LLM Systems Are Architecture, Not Prompts (Orca) (github.com)

A new AI framework called **Agent Skills Runtime**, part of the **ORCA (Open Cognitive Runtime Architecture)** specification, has been announced, emphasizing a shift from prompt-driven to execution-driven systems for AI agents. This framework allows developers to define agent capabilities as abstract contracts and use them to execute multi-step workflows as directed acyclic graphs (DAGs), enabling more deterministic and observable AI behavior. Agent Skills Runtime includes built-in safety features and cognitive state tracking, enhancing the reliability and control over agent executions. This development is significant for the AI/ML community as it formalizes a structured approach to agent reasoning. By decoupling reasoning from execution and establishing reusable capabilities, Agent Skills Runtime not only fosters safer interactions with APIs and databases but also streamlines the automation of workflows—potentially transforming how AI systems are built and integrated. With 141 pre-defined capabilities and support for multiple backend protocols, this architecture promotes efficiency in creating more robust AI applications without the need for API keys, making it accessible for rapid prototyping and development across various platforms.