/hyperpower:review

Split a diff into parts, review each part, then test every finding.

Reference

/hyperpower:review

When to use it

You have a diff too big to review well in one pass.

What you get

Findings marked CONFIRMED or PLAUSIBLE. A second agent tries to disprove each one before it reaches you.

How it works

It reviews the current diff, or a pull request when you pass its URL or number.

It splits the diff into parts that make sense on their own and reviews each part. Then a second agent tries to disprove every finding that came back.

Each finding gets one of two labels:

  • CONFIRMED. Traced end to end. The path from cause to effect is in the code.
  • PLAUSIBLE. The code alone cannot settle it. Deciding might take runtime context, or knowledge about production.

A PLAUSIBLE finding stays in the report even when nobody could reproduce it. A real bug that is hard to reproduce is still a real bug.

A forty-file diff reviewed in one go gets skimmed. Split into parts, each part gets a careful read.