What problem does it solve? Building self-improving agents requires implementing reinforcement learning from scratch, which is complex and error-prone. This Skill provides ready-to-use templates and APIs for training agents on logged experiences without writing RL algorithms manually. ## Core Features & Use Cases - Nine RL Algorithms: Decision Transformer, Q-Learning, SARSA, Actor-Critic, plus Active Learning, Adversarial Training, Curriculum Learning, Federated Learning, and Multi-Task Learning. - CLI Plugin Scaffolding: Generate learning plugins with npx agentdb create-plugin using interactive wizards or specific templates. - Experience-Based Training: Store state-action-reward experiences via the AgentDB adapter API and train models with configurable epochs, batch sizes, and validation splits. - Use Case: Train a game-playing agent by logging episode steps as experience patterns, then running adapter.train() to learn a policy, and retrieving suggested actions with confidence scores via retrieveWithReasoning. ## Quick Start Create a Decision Transformer learning plugin by running npx agentdb@latest create-plugin -t decision-transformer -n my-agent, then store experiences and call adapter.train() to begin training.
