automatic-stateful-prompt-improver

Prompt optimization automatically intercepts and refines your instructions using an MCP server and embedding-indexed history. It applies iterative improvement patterns like APE and OPRO to enhance multi-step reasoning tasks, reducing trial-and-error cycles and improving output quality across conversations.

You need a configured prompt-learning MCP server, a vector store like Qdrant or Chroma for embeddings, Redis for session metrics, and Claude Code integration. These components work together to maintain stateful learning and track performance data across optimization cycles.

Yes, prompt optimization works with system prompts, templates, and multi-step instructions. It's designed for complex reasoning tasks across domains and automatically learns from feedback to refine instructions, making it suitable for recurring or template-based workflows.

Stateful learning stores performance metrics and embedding-indexed history in persistent storage. Each optimization cycle converges toward better prompts using defined budgets, so future requests benefit from what worked in past tasks without restarting from scratch.

Optimization activates on requests containing keywords like 'optimize this prompt', 'improve this prompt', or 'prompt engineering', plus other complex instructions. Once triggered, it intercepts the prompt, refines it using the learning loop, and applies the improved version to your task.

Optimization runs within defined iteration budgets and convergence thresholds to keep improvements predictable and cost-controlled. Once thresholds are met, the system stops refining and returns the optimized prompt, preventing runaway iteration cycles.