copilot-review

You are reviewing and applying GitHub Copilot's code review feedback. Your job is to:

1. •Fetch Copilot's review comments
2. •Evaluate each suggestion critically
3. •Apply changes that make sense, skip those that don't
4. •Resolve the comment threads for applied changes

Step 1: Get PR details

Get the current PR number and repo info:

gh pr view --json number,url,headRefName

Extract the owner and repo from the URL (e.g. https://github.com/owner/repo/pull/123).

Step 2: Fetch Copilot review comments

Use GraphQL to get review threads with full context:

gh api graphql -f query='
  query {
    repository(owner: "OWNER", name: "REPO") {
      pullRequest(number: PR_NUMBER) {
        reviewThreads(first: 100) {
          nodes {
            id
            isResolved
            path
            line
            comments(first: 10) {
              nodes {
                author { login }
                body
                createdAt
              }
            }
          }
        }
      }
    }
  }
'

Filter for threads where the first comment author login contains "copilot" (case-insensitive).

If there are no Copilot comments:

• •Tell the user no Copilot feedback was found
• •Copilot may not have reviewed yet, or had no suggestions

Step 3: Parse Copilot's comment format

Copilot comments typically follow this structure:

In path/to/file.rb:

> @@ -21,6 +21,8 @@ def method_name
>    existing_line
>    another_line
+    new_suggested_line

[Explanation of the issue and reasoning]

Suggested change
-    old_code_line
+    new_code_line

For each comment, extract:

• •File path: from the "In path/to/file:" line or the thread's path field
• •Line number: from the thread's line field
• •Explanation: the prose explaining the issue
• •Suggested change: the diff block after "Suggested change" showing - (remove) and + (add) lines

Step 4: Evaluate each suggestion

Do not blindly apply all suggestions. For each one:

1. •Read the relevant file to understand the full context
2. •Understand Copilot's reasoning - what problem is it trying to solve?
3. •Evaluate critically:
   • •Does this suggestion actually improve the code?
   • •Is the reasoning sound given the full context?
   • •Are there side effects Copilot might have missed?
   • •Does it align with the codebase's patterns and conventions?

Apply the change if:

• •The reasoning is sound
• •The fix is correct
• •It improves code quality, performance, or correctness

Skip the change if:

• •The reasoning is flawed or based on incomplete context
• •The suggested code has bugs or issues
• •It conflicts with project conventions
• •The "problem" isn't actually a problem in this context

Step 5: Apply approved changes

For each suggestion you decide to apply:

1. •Read the file at the specified path
2. •Locate the code matching the - lines in the suggested change
3. •Replace it with the + lines
4. •Keep track of which thread IDs were addressed

Step 6: Resolve applied comment threads

For each comment thread where you applied the fix, resolve it:

gh api graphql -f query='
  mutation {
    resolveReviewThread(input: {threadId: "THREAD_NODE_ID"}) {
      thread { isResolved }
    }
  }
'

Use the thread id from the GraphQL query in Step 2.

Step 7: Report to the user

Present a clear summary:

Applied changes:

• •List each file modified
• •Briefly describe what was changed and why

Skipped suggestions:

• •List each suggestion that was not applied
• •Explain why (e.g. "Suggestion assumed X but the code actually does Y")

Next steps:

• •Changes have been applied but not committed
• •User should review with git diff
• •When satisfied, commit and push

Do not commit automatically. Let the user review first.