I think the way to go here is to leverage AI tools to reduce the time spent on PR reviews without compromising quality. Some PRs in this repo can grow large and/or require very specific domain knowledge, which makes it challenging for reviewers to tackle in a timely fashion.
At the moment, I can think of two main ways to address this:
- Train GitHub Copilot to provide better answers based on PRs, issues, and comments specific to our repository instead of mostly providing general coding advice. Maybe there's an equivalent to the
copilot-instructions.md file but for reviews only. Or maybe we can write skills for it. Keeping track of common patterns seen in the repo would also go a long way (memory).
- Some members of our team use skills, MCP servers, and CLI tools in conjunction with AI agents to reduce the cognitive burden of reviewing numerous changes while trying to understand the big picture in a given PR, reserving final say on what gets posted on GitHub after reviewing the AI-produced PR review. We should come up with a common toolset and configuration for our team members to run locally so we can run these reviews in interactive sessions on our preferred IDE/terminal/client.
I think the way to go here is to leverage AI tools to reduce the time spent on PR reviews without compromising quality. Some PRs in this repo can grow large and/or require very specific domain knowledge, which makes it challenging for reviewers to tackle in a timely fashion.
At the moment, I can think of two main ways to address this:
copilot-instructions.mdfile but for reviews only. Or maybe we can write skills for it. Keeping track of common patterns seen in the repo would also go a long way (memory).