AI coding assistants have become incredibly good at writing code.
But I kept running into the same problem...
When I asked an AI to upgrade a .NET application, the response usually looked something like this:
"Update the TargetFramework, update the NuGet packages, and build."
Anyone who has upgraded a real production application knows that's only a tiny part of the job.
A proper upgrade normally involves:
That's not a prompt.
That's an engineering workflow.
So I decided to build one.
CoreLogic AI Skills is an open-source library of reusable engineering workflows designed for AI coding agents.
Instead of writing the same prompts repeatedly, each workflow is packaged as a reusable Skill.
The goal is simple:
Build once. Use with any AI coding agent.
The library currently supports:
GitHub Repository:
👉 https://github.com/mrlokeshagarwal/corelogic-ai-skills
The first skill I created is called dotnet-upgrade.
Instead of blindly modifying a repository, it behaves more like an experienced senior engineer.
It first analyses the repository.
It looks for things like:
It then creates an upgrade plan.
Nothing is modified yet.
One design decision was very important to me.
The skill never starts changing code immediately.
Instead it presents something like this:
.NET Upgrade Assessment
Current framework:
.NET 6
Recommended target:
.NET 10 LTS
Projects:
6
Vulnerable packages:
4
Deprecated packages:
2
Docker changes required:
Yes
CI/CD changes required:
Yes
No files have been modified.
Approve this plan?
Only after explicit approval does the upgrade begin.
After approval the skill can:
The work isn't finished after changing code.
The skill validates the application by:
Because a successful build doesn't always mean a successful upgrade.
One thing I wanted to avoid was creating another repository full of prompts.
Instead, I built a small ecosystem.
The repository includes:
Adding a new skill shouldn't require changing the installer or packaging logic.
The repository also includes start-story.
It helps developers begin work from an Azure DevOps work item by:
Like dotnet-upgrade
, it includes approval checkpoints before potentially destructive actions.
Every AI coding tool has its own way of working.
Some support Skills.
Some support Rules.
Some support Instructions.
Some support AGENTS.md.
I wanted the workflow itself to remain portable while only the discovery mechanism changed.
That's why the repository includes adapters for multiple platforms rather than maintaining completely different implementations.
I'm planning to add more engineering-focused skills, including:
This is the first public beta.
If you use Claude Code, Cursor, OpenCode, Codex, ChatGPT—or any other AI coding assistant—I would really appreciate your feedback.
Which engineering workflow would you like to see turned into a reusable AI Skill?
⭐ GitHub Repository
https://github.com/mrlokeshagarwal/corelogic-ai-skills
Feedback, ideas and contributions are always welcome.