Claude Leak Confirms It: LLM Systems Are Architecture, Not Prompts (Orca)
The Agent Skills Runtime, based on the ORCA architecture, provides a deterministic, composable execution engine for AI agent workflows, emphasizing structured state, safety, and multi-protocol tool integration over prompt-driven models. It supports declarative workflows via YAML, ships with 141 Python baselines, and enables offline execution without API keys. The system is designed for reproducibility, observability, and safe interaction with external systems. It can integrate with frameworks like LangChain, CrewAI, and MCP-compatible clients.
Opening excerpt (first ~120 words) tap to expand
Agent Skills Runtime Agents should execute whenever possible. A deterministic, binding-driven execution engine for composable AI agent skills. Agent Skills Runtime lets you define agent capabilities as abstract contracts, wire them to any backend (Python, OpenAPI, MCP, OpenRPC), and execute multi-step workflows as declarative DAGs — with built-in safety gates, cognitive state tracking, and full observability. No API keys required. 141 capabilities ship with deterministic Python baselines. Install, run your first skill in under 3 minutes. ⚡ 30-second start Works on macOS, Linux, and Windows. No API key required. # 1 — clone and install git clone https://github.com/gfernandf/agent-skills.git cd agent-skills pip install -e .
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at GitHub.