Skip to content

PromptImprover Developer Documentation

Welcome to the PromptImprover developer documentation. PromptImprover is a Model Context Protocol (MCP)-first prompt governance layer designed for engineering workflows. It acts as an intelligent intermediary between an AI client and execution tools, adding repository-aware context, applying prompt refinement rules, and recording evidence to improve future runs.

Overview

PromptImprover demonstrates three key concepts: 1. Prompt governance before code execution: Ensuring prompts meet engineering standards before they are processed. 2. MCP-based integration: Utilizing a standardized protocol instead of editor-specific glue code. 3. Evidence-backed refinement: Using historical commits, tests, and repo context to continuously improve prompt quality.

Core Capabilities

  • Robust Commit Tracking: Dynamically fetches all commits since the last known SHA, ensuring a gapless historical record.
  • Agent Output Logging: Correlates execution outcomes with the prompts that generated them for continuous learning.
  • Predictive Learning: Derives autonomous engineering mandates and templates from historical actions.
  • RAG Snippets: Uses FlexSearch-based retrieval over the local codebase to inject relevant examples.
  • Persistent Memory: SQLite-backed storage for rules, patterns, and history.

Project Structure

  • universal-refiner/: The core MCP server, refinement engine, and history/learning layer.
  • gemini-extension/: Gemini-specific integration and CLI extension.
  • docs/: Architecture specifications and developer guides.

Getting Started

To dive deeper into how PromptImprover works internally, check out the Architecture Overview.