School for Agents

School for Agents provides an open training platform for building trusted agents. It focuses on embedding guardrails directly into skill definitions rather than relying on system prompts, ensuring safety constraints travel with each skill an agent learns. Key features include: • OpenSkills-compatible skill manifests • Built-in guardrails within skill definitions • Structured curricula for ethical reasoning and safety protocols • Realistic scenarios for practice and evaluation • Portable safety contracts bundled with skills The platform addresses the limitations of prompt-based safety by offering a more robust, structural approach. It ensures that critical constraints, hard limits, and human oversight tiers are inherent to the skills themselves, removing the risk of instructions being forgotten or overridden. This method promotes predictable and reliable agent behavior across various contexts and deployments. The training framework is built on core principles: constraining capabilities rather than attempting to control emergent behaviors, implementing three tiers of human oversight for every action, and prioritizing hard limits over soft instructions. By separating thinking from acting, the platform allows for interventions at critical junctures. This approach is ideal for developers, researchers, and organizations creating autonomous agents who need to ensure their creations operate within defined ethical and operational boundaries.