wtf is dotnet-modernize?

dotnet-modernize is a Python package that helps an AI coding agent upgrade an old .NET Framework project to a modern version of .NET. It follows an assess, plan, and build validated execute workflow similar to Visual Studio Copilot's Modernize agent, but the judgment heavy parts of the work are handled by an AI assistant like Claude Code reading a bundled instruction file, while this tool provides a command line helper for the mechanical, non reasoning parts of the process. The reasoning heavy tasks, such as reading a codebase, deciding the right upgrade strategy, and figuring out why a build broke, are deliberately left to the AI agent rather than reimplemented in Python. What this package does handle directly is installing its instructions into an agent's skill folder, scaffolding a new upgrade scenario, capturing a baseline build before any changes are made, building a knowledge graph of the codebase using a separate tool called graphify, and tracking progress by reading a checklist file. After installing with pip, pipx, or uv, you run it from inside your .NET solution folder. First you install the skill files into your agent's directory, then create an upgrade scenario with a target framework version, capture a baseline by running the existing build and tests without touching any code, and build the knowledge graph the agent will query. From there you hand the task to your AI agent with a plain instruction to upgrade the solution following the installed skill, and can check progress at any time with a status command that reads the checklist without needing the agent running. The package includes several other commands too, such as listing existing scenarios, archiving a finished one, querying the knowledge graph directly, and a doctor command that checks whether required tools like dotnet, git, and graphify are available. The author explains the design choice as drawing a clear line between things that need judgment, left to the AI agent, and things that are simple and testable, handled by deterministic code. It is licensed under MIT.