Better tools didn't improve Copilot code review. Rewriting their instructions did. 🧵

We swapped Copilot code review onto the same shared tools that power the Copilot CLI (grep, glob, view), expecting a clean upgrade. Instead, reviews did more work and caught fewer issues. 👀

@github This corrects the usual tooling story. Better instructions often beat another feature because they encode the team's review standard. The next step is versioning them like code and testing whether they reduce misses, noise, and reviewer rework.

@github i get it

@github I agree that instructions should be improved and also the default setup should be smart enough to call better tools on its own

@github Skill issue... but not on the AI's side. 😭

@github Turns out prompts are the real product.

@github Totally agree 💯

@github That's a prompt mismatch, not a tool failure.

10% off for the 4th of July