Автоматическое улучшение промптов с обучением

★ 6.8 · general

automatic-stateful-prompt-improver is a Claude Code skill that automatically intercepts user requests and optimizes them via the prompt-learning MCP server before generating a response. It applies APE, OPRO, and DSPy patterns, retrieves relevant history through embedding-indexed storage, and records success scores after each significant task so future optimizations benefit from past outcomes. Iteration count scales with complexity — 3–5 for simple tasks, up to 15–20 for agent pipelines — and stops automatically when improvement drops below 1% over three rounds or quality score exceeds 0.95. The skill is designed for multi-step reasoning, code generation, precision-critical domains, and reusable prompt templates; it deliberately skips optimization for simple questions, direct commands, and conversational replies.