Skip to content

feat: post a review summary even without findings and review large changed files around their diff - #18

Merged
hyochan merged 3 commits into
mainfrom
fix/review-summary-and-large-files
Nov 13, 2026
Merged

hyochan merged 3 commits into
mainfrom
fix/review-summary-and-large-files

Conversation

@hyochan

@hyochan hyochan commented Nov 13, 2026

Copy link
Copy Markdown
Member

Why

On hyodotdev/openiap#612, Gitdog posted nothing:

  • The first review covered 91 changed files and found no defects. Gitdog stays silent when there are no findings, so the PR looked unreviewed.
  • 57 of those files were never reviewed. Several of them were the PR's core sources (ExpoIapModule.kt, HybridRnIap.swift, useIAP.ts, the three OpenIapModule.kt variants). They were dropped because they are over 40 KB.

Changes

Always post a summary

  • A review with no findings now posts a summary review: "Gitdog found no supported defects in <sha>. N of M changed files were not fully reviewed. This is not a merge approval; tests were not run."
  • The summary is fixed text, so it is published without approval even when the repository requires approval. Reviews with findings still wait for approval as before.
  • Single reviews now record changedFiles and uncoveredFiles too, so their summary can report coverage.
  • Batched reviews now post their findings as inline comments, like single reviews, under a short summary body. Approvers still preview the full finding list in Tickets.

Review large changed files around their diff

  • A changed file over 40 KB, read at head, is now fetched up to 400 KB.
  • The reviewer sees only the lines within 80 lines of each hunk. If that window is too large, it narrows to 30 lines, then to the changed lines alone.
  • Every other line stays in place, either blank or as a marker line, so line numbers, anchors and suggestions are unchanged. If even the changed lines do not fit in 40 KB, the file is still reported as not reviewed.
  • Evidence may not quote the marker. The prompt tells the model not to infer omitted lines.
  • Publication re-reads the evidence the same way, so validation is deterministic for a given head commit.
  • Base reads and unchanged files keep the 40 KB limit.

Tests

  • bun run lint, typecheck, format:check, build and all 1899 tests pass.
  • New tests cover:
    • window layout and line numbers, and progressive narrowing;
    • the 400 KB refusal, and unchanged/removed files keeping the 40 KB limit;
    • rejecting a quote of the marker;
    • the summary body with coverage;
    • publication of a zero-finding review without approval or inline comments;
    • inline comments on batched reviews.

🤖 Generated with Claude Code

…anged files around their diff

A review that found nothing used to post nothing, so a PR looked unreviewed and a review that skipped most files read as clean. It now always posts a summary with how many changed files were not fully reviewed; the fixed no-findings text needs no approval. Batched reviews post their findings as inline comments under that summary.

Changed files over 40 KB used to be skipped entirely. They are now read up to 400 KB and shown around their diff with the original line numbers, and evidence may not quote the omitted-lines marker.
@hyochan
hyochan merged commit f163ebe into main Nov 13, 2026
7 checks passed
@hyochan
hyochan deleted the fix/review-summary-and-large-files branch November 13, 2026 16:15
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