Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
47 changes: 47 additions & 0 deletions .github/copilot-instructions.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,47 @@
# Copilot code review instructions

How reviews on this repository should be conducted. These govern the review
process itself, not what counts as good code.

## Report everything in one pass

Report every finding you have in a single review. Do not hold findings back for a
later round.

- Review the complete diff — every changed file, every changed hunk. Do not stop
partway through.
- Finding several problems early is not a reason to end the review. It is a reason
to keep going.
- Do not sample. If a file has twelve problems, report twelve.
- Comment on every instance of a repeated problem, not just the first. If the same
mistake appears in six places, mark all six — the author fixes what is marked, so
"and similar issues elsewhere" leaves five defects in the branch.

## Include minor and uncertain findings

Do not filter down to only the findings you are most confident about.

- Report low-severity and minor issues alongside significant ones.
- Report a finding you are less than certain about, and say plainly what you are
unsure of. A finding the author dismisses in ten seconds costs far less than an
entire extra review cycle.
- Label each finding's severity so the author can triage quickly instead of having
to weigh every comment equally.

## Converge in as few rounds as possible

Every additional review round costs the author a full cycle. Target:

- **One round** — ideal. Everything surfaced on the first review.
- **Two or three rounds** — acceptable.
- **Five rounds** — the absolute ceiling.

A later round should raise only problems introduced by the previous round's fixes.
A problem that was present in the original diff and went unmentioned until round
three is a review failure, not a thorough review.

## Do not trade thoroughness for brevity

None of the above is a reason to scrutinize less carefully. The goal is more
findings per review, not faster reviews. Depth of analysis stays the same; what
changes is that all of it arrives at once.
Loading
Loading