Claude Leak Confirms It: LLM Systems Are Architecture, Not Prompts (Orca)
Agent Skills Runtime is a new system that enables AI agents to execute structured workflows through deterministic, contract-driven processes rather than relying on prompts. It supports multi-step workflows as declarative DAGs with built-in safety, cognitive state tracking, and full observability, operating offline without API keys. The framework is based on ORCA, an emerging standard for structured agent execution that emphasizes explicit reasoning and reusable capabilities.
- ▪Agent Skills Runtime allows AI agents to execute composable skills through deterministic workflows instead of prompt-based reasoning.
- ▪It includes 159 pre-built capabilities with Python baselines and supports integration with multiple protocols including OpenAPI, MCP, and OpenRPC.
- ▪The system operates offline, requires no API key, and provides safety features such as trust levels, confirmation prompts, and side-effect tracking.
- ▪Workflows are defined in declarative YAML, enabling reproducibility, observability, and auditability via OpenTelemetry and Prometheus.
- ▪ORCA, the underlying architecture, externalizes agent cognition into structured state, capabilities as contracts, and execution graphs for safer, more transparent AI agent behavior.
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. 159 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.