Automated GitHub AI Review Agent using C# Semantic Kernel

[rwjdk]

If you are interested in samples of SK usage for your blog, we used SK to fully automate a AI Review Agent on Github using SK to do first-pass reviews before a human do the "real" review and would be happy to do a write-up

• A GitHub app cofigured to send webhook events to a minimal API to automate the triggering of Reviews
• SK to do a review based on PR diff and Task-description that is related to the PR (aka what the user was intended to do) [gpt4o or o3-mini]
  • SK use structured output to give review back + what additional files might benefit review
  • If extra files could benefit; on the fly retrieve them from GitHub Repo and let SK do a second pass review including the additional for better quality
• Post finding as GitHub PR Comment for the Developer to see before assigning seconday human reviewer

Benefit compared to GitHub Copilot:

• You are 100% control of models, prompts and flows of how things should be configured and look visually
• Your are in control where code is shared with services (if sensitive)
• Cheap as it is 'token usage based' instead of subscription based

[sophialagerkranspandey]

Thanks for reaching out! @rwjdk I'll email you to connect further on this.