Обновление скилла до формата Decision Theory v5

ai-tooling

skill-upgrader is a Claude Code skill that upgrades any SKILL.md to the Decision Theory v5 Hybrid format by spawning 4 parallel agents via the Task tool. Each agent queries its own book through Ragie RAG: Agent 1 (LaValle "Planning Algorithms") identifies MDP structure — states, actions, and transitions; Agent 2 (Sutton & Barto "Reinforcement Learning") designs the policy, termination conditions, and reward function; Agent 3 (Blackburn "Modal Logic") extracts temporal, epistemic, and deontic constraints; Agent 4 (Huth & Ryan "Logic in Computer Science") verifies safety and liveness properties. All findings are written to a shared blackboard, and the skill then synthesizes the final SKILL-upgraded.md. It requires Ragie with decision-theory and modal-logic partitions pre-indexed, and is aimed at teams formalizing agentic skills using MDP frameworks and modal logic.