Skip to content

v1.4.1 - Reviews verified against the PR's real code

Choose a tag to compare

View all tags
@tms-minhtang1 tms-minhtang1 released this 21 Aug 16:34
· 26 commits to main since this release
v1.4.1
228cb78
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Update, if the plugin is already installed:

/plugin marketplace update open-pr
/plugin update open-pr@open-pr

then /reload-plugins. Cursor, Codex, Gemini CLI, Antigravity:

curl -fsSL https://raw.githubusercontent.com/TOMOSIA-VIETNAM/open-pr/main/install.sh | bash -s -- --update

Existing repos need no config migration — schema_version is unchanged.

Improved

  • Every review is verified to be looking at the PR's real code: the worktree — and a submodule's checkout on a bump — must match the PR's head commit before any finding is written, and a failed checkout stops the run instead of reviewing the wrong tree.
  • Reviewing a PR whose branch is checked out in your own clone now works — the checkout fetches the PR ref directly instead of gh pr checkout, which exited 128 there.
  • Line comments land on the line they describe: a LINE finding must sit inside a diff hunk, and its line number is confirmed against the real file (the merge-base blob for the old side) before posting.
  • A push landing mid-review no longer wastes the run: the review posts anchored to the commit that was read, with a note that a newer commit is unreviewed.
  • Submodule reviews read the submodule's own tree for line checks, old-finding comparison and merge base — not the parent repo's file at the same path — and anchor their post to the commit they reviewed.
  • The overview's assessment is capped at three sentences, one conclusion each.
  • /open-pr:feedback writes the tracker's real form headings, so the issue body matches the forms exactly.
  • A post failure on your own PR is no longer declared unworkaroundable, and a rejected commit_id is explained as a force-push with the fix (call the review again).
  • Context cost: mean +4.4% (+403 tokens per run), all of it spent on the verification gates above; fix, upgrade and clean are unchanged.
Assets 2
Loading