AI code review that lands on the PR, not in your terminal — GitHub, GitLab, Bitbucket. Learns each repo's conventions: 1 review, 1 fix commit, replies in thread. Claude Code · Cursor · Codex · Gemini CLI · Antigravity.
# Add to your Claude Code skills
git clone https://github.com/TOMOSIA-VIETNAM/open-prLast scanned: 9/17/2026
{
"issues": [
{
"file": "README.md",
"line": 81,
"type": "remote-install",
"message": "Install command (remote install script piped to a shell — review the source before running): \"curl -fsSL https://raw.githubusercontent.com/TOMOSIA-VIETNAM/open-pr/main/instal\"",
"severity": "low"
}
],
"status": "PASSED",
"scannedAt": "2026-09-17T09:15:01.088Z",
"npmAuditRan": true,
"pipAuditRan": true,
"promptInjectionRan": true
}See how open-pr compares with popular alternatives.
open-pr is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by TOMOSIA-VIETNAM. AI code review that lands on the PR, not in your terminal — GitHub, GitLab, Bitbucket. Learns each repo's conventions: 1 review, 1 fix commit, replies in thread. Claude Code · Cursor · Codex · Gemini CLI · Antigravity. It has 44 GitHub stars.
Yes. open-pr passed SkillsLLM's automated security scan — a dependency vulnerability audit plus prompt-injection heuristics — with no high-severity issues. You can read the full report in the Security Report section on this page.
Clone the repository with "git clone https://github.com/TOMOSIA-VIETNAM/open-pr" and add it to your Claude Code skills directory (see the Installation section above).
open-pr is primarily written in Python. It is open-source under TOMOSIA-VIETNAM on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh open-pr against similar tools.
No comments yet. Be the first to share your thoughts!
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
AI coding made PRs faster. But review didn't get faster.
open-pr runs that first review round for you — on the PR, not on your laptop. Anyone who opens the PR sees the same feedback.
One run produces three parts that belong together: an overview, line comments (with suggested changes), and a reply after /open-pr:fix has pushed. — See the demo
/open-pr:fix is disciplined — exactly 1 commit, no force-push, a reply on every threadopen-pr service — MIT-licensed, no open-pr server or bot account; it runs in the agent CLI you already haveopen-pr is listed in Awesome AI Plugins — the cross-platform catalog curated by Hashgraph Online — under Community Plugins → Development & Workflow, and it is ingested by the HOL Plugin Registry, which scans every listed project and publishes a trust score.
The listing is gated on that scan: score ≥ 80, with no high or critical finding. The catalog runs the scanner itself against this repository's default branch, so the result is theirs rather than ours. A scan is a trust signal, not a safety guarantee.
1. A vendor CLI, logged in. The plugin carries no credential of its own — it reads the PR and posts the review through your account:
# GitHub
brew install gh # or https://cli.github.com/
gh auth login # GitHub.com → HTTPS → Login with a web browser
gh auth status # must say "Logged in to github.com as <you>"
GitLab: brew install glab && glab auth login --hostname gitlab.com. Bitbucket ships no CLI — it reads BITBUCKET_EMAIL + BITBUCKET_API_TOKEN from the environment. Minimum permissions and how to check them: Getting a token per vendor.
2. The plugin. Claude Code:
/plugin marketplace add TOMOSIA-VIETNAM/open-pr
/plugin install open-pr@open-pr
Install for Cursor, Codex, Gemini CLI, Antigravity:
curl -fsSL https://raw.githubusercontent.com/TOMOSIA-VIETNAM/open-pr/main/install.sh | bash
Full guide: Install · Getting a token per vendor.
PR URL formats: GitHub .../pull/<n> · GitLab .../-/merge_requests/<n> (self-hosted included) · Bitbucket Cloud .../pull-requests/<n>.
In the age of AI coding, PRs ship far faster than they get reviewed. The bottleneck is no longer coding — it's review. Reviewers have to check the project's conventions / security / performance and cover business logic — and that doesn't scale well as PR volume grows.
A local review is hard to trust. Anyone can say "I already reviewed it". So open-pr moves that step to remote for transparency — comments sit on the PR, and anyone who opens it can see them.
[!NOTE] Review rounds (a suggestion for the team):
- Round 1 — Dev runs AI review on the PR themselves. No review comments yet → reviewer sends it back, without touching it.
- Round 2 — Reviewer runs it again (AI). Clean → LGTM.
- Round 3 — Reviewer reviews the domain part.
[!IMPORTANT] AI lightens the process load, but final responsibility is still yours.
Many review skills are just a SKILL.md description. Each run comes out differently — different wording, different strictness, easy to drift from the project's conventions.
| Common with a generic skill | With open-pr |
|---|---|
| Advice stays at generic rules, off the project | Reads README / CLAUDE.md / AGENTS.md / docs / wiki; team rules beat generic rules |
| You remind it once, next time it slips again | Mentions in chat → asks to write into the repo's memory → next run applies it |
| Told to fix → fixes per the comment — even a wrong comment → correct code becomes wrong | /open-pr:fix weighs whether a comment is sound; if not → reply + evidence, no code change |
| Fixes arrive as commit spam, amends, force-pushes, no replies | Exactly 1 commit per fix, no history rewriting, a reply per comment after push |
[!TIP] The part worth keeping: no matter when you run it, the procedure is the same — bootstrap conventions, pick the output language from the repo, then remember what the team has reminded. Not one AI voice today and another tomorrow.
flowchart LR
A[New PR] --> B["Round 1 · /open-pr:review"]
B --> C{Review on remote?}
C -- not yet --> D[Reviewer sends it back]
C -- yes --> E[Dev fix / /open-pr:fix]
E --> F["Round 2 · review again"]
F --> G{Clean?}
G -- yes --> H[LGTM]
G -- not yet --> E
H --> I[Round 3 · human domain review]
Details on re-review, worktrees, and the guard before fix: Re-review / fix flow.
| Command | What it does |
|---|---|
/open-pr:review <PR_URL> |
Posts exactly 1 review. No code edits, no close, no merge. First run in a repo also sets it up |
/open-pr:fix <PR_URL> |
Reads findings → weighs right/wrong → fixes → 1 commit → replies. 🔵 / 📝 always ask first |
/open-pr:upgrade |
Brings local config up to the current schema — summarises, then asks; nothing written until you agree |
/open-pr:clean |
Removes worktrees that review checked out (asks first). Memory / settings untouched |
/open-pr:feedback |
Reports a problem with this plugin on its issue tracker — stripped of anything identifying your repo, and shown to you before it is posted |
[!WARNING]
fixedits real code in the repo (or the review worktree). Run it only when you deliberately want it to handle the comments.
Full configuration: Configuration.
Criteria in detail and priority when they conflict: What it reviews.
Mean tokens per run — covering both happy-case and