Skip to content

fix: one review result, and reviews that read the conversation - #34

Merged
hyochan merged 3 commits into
mainfrom
fix/one-review-result
Nov 18, 2026
Merged

hyochan merged 3 commits into
mainfrom
fix/one-review-result

Conversation

@hyochan

@hyochan hyochan commented Nov 18, 2026

Copy link
Copy Markdown
Member

Why

On a pull request in a connected repository:

  • The maintainer's review asked for two changes before merging, and another review bot had an unresolved Major thread. Gitdog only looked at the code, found nothing, and approved.
  • Gitdog also left a "Reviewed · see the review" status comment next to the review itself, so the result showed up twice.

What changes

  • One result. Once a review is posted, its status comment is deleted (deleteIssueComment; a 404 counts as done). The comment still shows reviewing, waiting for approval, failures and no-change results, where no review exists.
  • Approval respects other reviewers. A clean review no longer approves while:
    • any review thread is unresolved, whoever opened it;
    • a reviewer's latest decision is "changes requested";
    • an open request is listed.
      A review that lists open requests also withdraws an earlier Gitdog approval, as findings do.
  • Reviews read the conversation. A single review reads the PR's comments (newest page), review summaries and the last 100 unresolved threads.
    • People's entries come first, then bots', without Gitdog's own posts or command lines.
    • Each entry is capped at 1,200 characters, 8,000 in total, and passed as untrusted data.
    • The reviewer returns openRequests, listed under "Not yet addressed from the discussion". Such a review waits for approval like one with findings.
    • An open request counts only if it is one line of plain text with no links or cross-references, under a login that took part, sharing wording with what that person wrote. Mentions are quieted.
    • Batched (large) reviews don't read the discussion, but open threads and requested changes still block their approval.

Production impact

  • Up to three extra GitHub reads per single review.
  • The prompt and schema grow by about 0.5K, and the discussion adds up to about 8K to the main review's context.
  • Repositories where review threads often stay unresolved will see fewer Gitdog approvals, which is intended.
  • No schema migration: proposedChanges is untyped.

Verification

  • bun run lint, typecheck, build, test (2090 tests), format:check, check:api, typecheck:cli, test:cli
  • Independent Opus and Fable reviews found no blockers. Their should-fix items are addressed:
    • approval withdrawal;
    • attribution and formatting of requests;
    • newest-page reads;
    • short requests;
    • docs.
      Fable confirmed the head clean.

🤖 Generated with Claude Code

A posted review is the result, so its status comment is now deleted
instead of repeating it as "Reviewed". A single review also reads the
pull request conversation (comments, review summaries and unresolved
threads, bounded and without Gitdog's own posts) and lists each change a
reviewer asked for that the code does not make yet. A clean review no
longer approves while such a request, an unresolved thread from anyone,
or a reviewer's standing request for changes remains.
…eir own

A review that lists open requests now withdraws an earlier Gitdog
approval like one with findings. An open request counts only if it
shares wording with what that reviewer wrote, and only as one line of
plain text without links or cross-references. The discussion reads the
newest conversation page and the last hundred threads, puts people ahead
of bots, keeps short requests, and drops only the command line of a
comment. Publish skips thread reads that cannot change the outcome, and
verification-only reviews do not read the discussion.
Open requests are asked for in the reviewer's own words so the wording
check holds, and one failed GitHub read no longer drops the other parts
of the discussion.
@gitdogai

gitdogai Bot commented Nov 18, 2026 •

Copy link
Copy Markdown

Review cancelled for 541ce1d

The pull request changed or was closed before the review finished.

@hyochan hyochan left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

4 issues found while reviewing this change, all fixed in 50f8148

Not a merge approval. Tests were not run. Gitdog

Comment thread convex/pullRequests/internalAction.ts Resolved
Comment thread convex/pullRequests/review.ts Resolved
Comment thread convex/pullRequests/review.ts Resolved
Comment thread convex/pullRequests/github.ts Resolved
@github-actions

Copy link
Copy Markdown

Visit the preview URL for this PR (updated for commit 541ce1d):

https://codeyou-df251--pr34-fix-one-review-resul-4fw9npuu.web.app

(expires Wed, 25 Nov 2026 05:46:05 GMT)

🔥 via Firebase Hosting GitHub Action 🌎

Sign: 844dad9c62537d028cabc21008e4e2e4031ce4e5

@hyochan
hyochan merged commit 23bcd96 into main Nov 18, 2026
9 checks passed
@hyochan
hyochan deleted the fix/one-review-result branch November 18, 2026 05:47
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant