Skill: Open Code Review
by Alibaba Group
Alibaba's open-source AI code review CLI as an agent skill: line-level comments on diffs, branches and commits, with optional auto-fix.
Open source pre-push AI code review for Claude Code and Codex: scanners plus a separate reviewer that verifies every finding.
One command for Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf and dozens of other agents. Needs Node.js; the skill is fetched straight from openqodex/openqodex.
$ npx skills add openqodex/openqodex Run it in a terminal. The CLI finds the agents installed on your machine, asks which ones to add the skill to, and whether to install it into the folder you ran it from or for all your projects. It also lists the skills in the repository so you can pick OpenQodex.
SKILL.md.~/.claude/skills/openqodex/, where
Claude Code picks it up in every project.No scripts or references next to it? Saving the SKILL.md alone as ~/.claude/skills/openqodex/SKILL.md works too.
SKILL.md inside its top folder.Paste this into an agent that can run commands — Claude Code, Codex, Cursor — and it installs the skill itself.
Install the agent skill "OpenQodex" from https://github.com/openqodex/openqodex. If Node.js is available, run `npx skills add openqodex/openqodex -g -y -a <your agent id>`, where the id is claude-code, codex, cursor, github-copilot, gemini-cli, windsurf or opencode. Otherwise copy that folder — SKILL.md and every file beside it — into your personal skills directory (for Claude Code: ~/.claude/skills/openqodex/). Then tell me the skill is installed and what it does. Instructions from OpenQodex, as published with the skill.
Any agent
npx openqodex initThen tell your agent: "review my change with openqodex".
OpenQodex reviews your change before you push it. One command, openqodex review, works out what you are about to ship (unpushed commits plus uncommitted edits), runs the scanners that match the changed files (SAST, secrets, dependencies, lint) and keeps only findings on the lines you actually touched.
A second, independent reviewer process, started as a separate Claude Code or Codex session on a frozen copy of the change, then checks every scanner finding and is given every changed line. OpenQodex validates that answer with scripts and writes a single report as HTML, Markdown, JSON and SARIF, so each finding appears under its line of code. You choose which findings your agent fixes.
It needs Node 22+, git, and a logged-in Claude Code or Codex; no extra API key, account or server. npx openqodex init detects Claude Code, Cursor, Codex CLI and Cline and installs the skill, and a lighter skill-only install is available. Apache-2.0 licensed; runs on macOS and Linux (WSL on Windows).
The instructions your agent loads, as published by OpenQodex.
<picture>
<source media="(prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/openqodex/openqodex/main/.github/assets/readme-banner-dark.png">
<img alt="OpenQodex" src="https://raw.githubusercontent.com/openqodex/openqodex/main/.github/assets/readme-banner-light.png" width="1280">
</picture>



OpenQodex is open source AI code review for Claude Code and Codex. It runs before you push, from your coding agent or your terminal. One command, openqodex review, works out your change: the commits not yet pushed plus everything uncommitted. It runs the scanners that fit the changed files and keeps only findings on the lines you changed. Then it starts its own reviewer, a separate Claude Code or Codex process that reads a frozen copy of the change. The reviewer checks every scanner finding and is given every changed line. OpenQodex checks its answer with scripts, writes one report, and prints a short receipt: the verdict, one line per finding and the path of report.html, a local page that shows each finding under its line of code. You choose which findings your agent fixes. It needs Claude Code or Codex installed and logged in, and no other key, account or server.
For humans, in your terminal:
npx openqodex initinit finds Claude Code, Cursor, Codex CLI and Cline on your machine. It prints every file it will write and asks once. Then it names the reviewer it found, or what to fix, and reviews your change, or asks what to review when there is none. After that, say to your agent "review my change with openqodex", or run ~/.openqodex/bin/openqodex review yourself: init prints that full path, since an npx install puts no openqodex on your PATH.
For agents, the same install, run by the agent for itself with no question:
npx -y openqodex@0.11.0 init --yes --agent <host><host> is claude-code, codex, cursor or cline. Or paste this prompt into your agent:
Install OpenQodex for yourself with `npx -y openqodex@0.11.0 init --yes --agent <host>`, where <host> is the agent you are: claude-code, codex, cursor or cline. Run it from this repository and allow it up to ten minutes: when a reviewer can start, it ends with a review of my current change.
Then tell me the verdict and the findings, or what its last lines say is missing.Codex runs commands in a sandbox that by default cannot write outside the project or reach the network: from Codex, run the line in your own terminal instead.
The skill alone, with no push check, launcher or scanner download: npx skills add openqodex/openqodex -g. A later init replaces it with the skill it keeps up to date.
OpenQodex needs Node 22 or newer and git. It runs on macOS and Linux. On Windows, use WSL.
<!-- recording: added before launch -->
Five commands: init, review, update, trust and graph. The commands hooks and agents call are listed in [docs/plumbing.md](docs/plumbing.md).
openqodex review runs the whole review in one command: a frozen copy of the change, the scanners, the code graph, a reviewer process OpenQodex starts, script checks of its answer, and one report in report.html, report.md, report.json and report.sarif.report.html and report.md. --format markdown, json or sarif still prints the whole report.report.html is one file on your disk: each changed file as a diff, each finding under its line, then the coverage, the scanners and the blast radius. It runs no script, loads nothing, escapes every string and redacts the secrets the scanners found.openqodex findings 1,3 prints the named findings in full for it.openqodex review --all reviews the whole repository. openqodex review <branch> and openqodex review '#42' review a branch or a pull request that is not your current work. OpenQodex fetches it, checks it out in a temporary folder and reviews what it added since it left its base.claude -p) or Codex (codex exec). auto picks the agent you run the command from, then Claude Code, then Codex; --reviewer or reviewer: in ~/.openqodex/config.yaml picks one.~/.codex/AGENTS.md, and its event stream does not show every command, so the report says its reads were not recorded. [docs/internal-reviewer-drivers.md](docs/internal-reviewer-drivers.md) gives the tests.# nosec, and a changed scanner settings file are shown, since the scanner then stays silent: the reviewer checks each one, and a scan counts it as a minor finding.openqodex trust..openqodex/config.yaml sets review.block_on_severity.openqodex scan) without one. A pre-commit hook that runs the scanners only; it is not a review.npx openqodex demo builds a small repo with planted bugs and scans it.openqodex graph callers <symbol> and the other graph questions answer from the code graph, with the evidence for each item and a note when the list may be short. init registers the same questions as an MCP server (openqodex mcp, a tool server your agent starts) with each agent it installs into. [docs/graph.md](docs/graph.md) lists the questions.review prints "Full review unavailable", says what is missing, saves the unchecked scanner findings to a file it names, and names the command with which the agent you are in reviews the change itself (review --agent). That report says which agent reviewed.cursor-agent has no way to limit its tools to reading or to skip your rules and settings.openqodex review runs inside Codex's own sandbox: a second Codex does not start there. review then prints "Full review unavailable" and the review --agent command.Scanners download on first use into ~/.openqodex/tools/. A review downloads only the scanners its changed files call for. init and doctor --install download the ones your repository's files call for and print why for each one, such as brakeman: Rails app in backend/. The table below gives each download size.
Installed scanners take more disk than their downloads. The sixteen scanners the demo needs take about 1 GB of disk on an Apple Silicon Mac, measured on 2026-10-08, besides uv's download cache in ~/.openqodex/cache/uv. semgrep with its Python takes about 250 MB of that, Checkov about 175 MB and trivy about 160 MB.
A scanner install that takes longer than 45 seconds keeps going in the background. The report lists that scanner as installing. The scanner joins the next run. The review init ends with waits up to two minutes, since init has just started the downloads. To install what a repository needs up front, run npx openqodex doctor --install inside it. --all-scanners installs every scanner.
One measured first run, on 2026-10-02, when the demo needed eight scanners: an Apple Silicon Mac, an empty tool folder, a line of 2 MB per second. The first openqodex demo printed its report in under a minute. That report held the scanners that had finished installing and listed the rest as installing. The next scan included all eight scanners. Your times depend on your line.
OpenQodex does not install language runtimes. brakeman needs Ruby 3.0 or newer and rubocop Ruby 2.7 or newer. golangci-lint needs Go. cargo-deny needs Cargo (Rust) and the project's crates in your Cargo cache. Without them, the report lists those scanners as not installed, with the reason.
| Scanner | Version | Runs when the change holds | Needs | Download (Apple Silicon, Linux x64) |
|---|---|---|---|---|
| semgrep | 1.94.0 | any file | Python 3.11, downloaded through uv | about 86 MB with bandit, measured on Apple Silicon |
| gitleaks | 8.21.2 | any file | nothing | 2.9 MB, 3.0 MB |
| bandit | 1.9.4 | .py, .pyi | the same Python as semgrep | included with semgrep |
| ruff | 0.8.4 | .py, .pyi | nothing | 9.9 MB, 11.2 MB |
| oxlint | 1.86.0 | .js, .jsx, .ts, .tsx, .mjs, .cjs, .mts, .cts | nothing | 5.3 MB, 6.1 MB |
| osv-scanner | 2.6.0 | a lockfile, such as package-lock.json, bun.lock, uv.lock or go.mod | network access to osv.dev | 52.6 MB, 54.9 MB |
| actionlint | 1.7.7 | .github/workflows/*.yml | nothing | 2.0 MB, 2.1 MB |
| hadolint | 2.15.1 | a Dockerfile | nothing | 102.6 MB, 55.7 MB |
| shellcheck | 0.10.0 | .sh, .bash, or a file with no extension whose first line (#!) names sh, bash, dash or ksh | xz to unpack | 7.2 MB, 2.4 MB |
| golangci-lint | 2.12.2 | .go | Go | 14.4 MB, 15.0 MB |
| brakeman | 6.2.1 | a Ruby or Rails file in a Rails app: a folder whose Gemfile or Gemfile.lock names rails and that holds config/application.rb or bin/rails | Ruby 3.0 or newer; see its licence below | from RubyGems, not measured |
| rubocop | 1.69.2 | .rb, .rake, .gemspec, Rakefile | Ruby 2.7 or newer | from RubyGems, not measured |
| sqllint | built in | .sql | nothing, it runs inside OpenQodex | none |
| squawk | 2.66.0 | .sql | nothing | 16.9 MB, 25.7 MB |
| SQLFluff | 4.3.0 | .sql | the same Python as semgrep | about 4.6 MB, 6.0 MB of packages |
| zizmor | 1.30.1 | a GitHub workflow, an action's action.yml, or .github/dependabot.yml | nothing | 8.4 MB, 9.2 MB |
| trivy | 0.75.0 | Terraform, a Kubernetes manifest or a CloudFormation template (its config checks only) | nothing | 49.0 MB, 51.7 MB |
| Checkov | 3.3.22 | the same files as trivy | the same Python as semgrep | about 170 MB installed on Apple Silicon |
| TFLint | 0.64.0 | .tf, .tf.json | nothing | 16.3 MB, 17.1 MB |
| kube-linter | 0.8.3 | a .yaml or .yml file that holds a Kubernetes object | nothing | 15.3 MB, 16.5 MB |
| kubeconform | 0.8.0 | the same files as kube-linter | network access to raw.githubusercontent.com for schemas | 7.3 MB, 7.5 MB |
| cargo-deny | 0.20.2 | Cargo.lock | Cargo (Rust), the project's crates in your Cargo cache, and network access to github.com for the RustSec advisory database | 4.5 MB, 4.9 MB |
[docs/scanners.md](docs/scanners.md) lists every file each scanner reads and what each one sends.
brakeman's licence is the Brakeman Public Use License, which is not an open source licence. It runs only for a Rails app. OpenQodex does not bundle brakeman. It downloads brakeman at run time onto your machine. scanners.disable: [brakeman] switches it off.
Add a scanner by its GitHub link in your repo's .openqodex/config.yaml:
scanners:
custom:
- source: https://github.com/aquasecurity/trivy
run: trivy config --disable-telemetry --skip-version-check --skip-check-update --format sarif --output {report} {target}A custom scanner is a command that runs on your machine. It never runs until you approve it:
npx openqodex trusttrust picks the release asset for your machine and downloads it. It shows the version, the asset, its sha256 and the run line, then asks yes or no. An edited entry needs a new approval. [docs/custom-scanners.md](docs/custom-scanners.md) explains each step.
p/default, p/security-audit, p/secrets), fetched from the Semgrep registry on each run.reviewer_web: off in ~/.openqodex/config.yaml removes the web tools.~/.codex/AGENTS.md and the output of the commands it runs in the copy of the change, to the model your Codex login uses. It can also use Codex's cached web search unless reviewer_web: off is set; its commands get no network either way.openqodex trust reads the custom scanner's release from the GitHub API and downloads it.openqodex review <branch> or review '#<number>' fetches that branch or pull request from your remote with git, and asks gh for the pull request's base when gh is installed.init, a version check at most once a day: the openqodex release list from registry.npmjs.org, and for a newer release its tarball and signed build record. It sends no code and nothing about you.--offline skips osv-scanner and semgrep and turns scanner downloads, the version check, and the fetch and gh call of a branch or pull request review off.
The built-in scanners send no code anywhere. The reviewer's model sees the brief and what the reviewer reads, as with any Claude Code or Codex session. A custom scanner you approved does whatever its own command does. [docs/security.md](docs/security.md) gives the full list.
An install made with npx openqodex init from 0.3.0 on keeps itself up to date. At most once a day, after a review, a scan or a push check, a background process looks for a new release. The command never waits for it. A release is installed only when it is at least 24 hours old and its signed build record (npm provenance) shows it was built by this repository's release workflow. It goes into a folder of its own beside the version you run, and the switch is one rename of a small file, so a failed or interrupted update leaves the working version in place.
An update writes only that folder, the file that names the active version and its own state. It never writes your settings, a repository or the files init wrote for your agents (the skill, the rules, the instruction lines, the Claude Code hook and permission rules). In user scope those files hold no procedure: the skill asks the launcher for the procedure of the version that runs. A release that changes how agents run a review, or the config format, is not installed in the background: the next command says so, and openqodex update installs it. When files init wrote are from an older version, the next command after an update says how many and that init refreshes them; init keeps every file you edited. That command also says which version it moved to, and names any change in what leaves your machine, what blocks a push or who reviews. openqodex update --rollback goes back, and that release is never installed again; updates stay on, so the next one comes (going back to 0.8.1 or earlier turns them off instead).
Turn it off with openqodex update --off, update: off in ~/.openqodex/config.yaml or OPENQODEX_AUTO_UPDATE=0. It is also off with --offline and when CI is set, and paused while that file holds a key OpenQodex does not know.
These do not update: the files init wrote for your agents until init runs again, files committed with init --project, the review section init adds to a repository's CLAUDE.md and AGENTS.md, the skill from npx skills add until the next init replaces it, the GitHub Action pin, and machines that are offline or stop background processes. An active install is usually one to two days behind a release. An install made with any earlier version needs one npx openqodex init to start updating. An install of 0.8.1 or earlier does not know about agent contracts yet: its next update installs the newest release whatever it changes, and the command after it names the files init would refresh.
| Package | What it is |
|---|---|
| `openqodex` | One package. It holds the CLI as one bundled file with no runtime dependencies. The skill, the agent templates, the docs, the review patterns and the demo ship as separate files beside it. |
@openqodex/core and @openqodex/scanners are internal workspace packages. The CLI bundles them, and they are not published.
The docs ship inside the package. npx openqodex guide <topic> prints a page offline.
None. OpenQodex sends no usage data. semgrep runs with its own metrics switched off. See [docs/telemetry.md](docs/telemetry.md) and the privacy policy, [docs/privacy.md](docs/privacy.md).
Report a vulnerability through GitHub's private vulnerability reporting on this repo. Never open a public issue for one. See SECURITY.md.
OpenQodex is new and on the way to 1.0. Commands, flags and the config file can change between minor releases. CHANGELOG.md records every change.

OpenQodex is made by Qodex, which also runs a hosted review on every pull request.
Licensed under [Apache 2.0](LICENSE). See [NOTICE](NOTICE) for the scanners' licences.
by Alibaba Group
Alibaba's open-source AI code review CLI as an agent skill: line-level comments on diffs, branches and commits, with optional auto-fix.
by Addy Osmani
Reduces accumulated complexity without changing behaviour, using Chesterton's Fence to avoid deleting the thing that was load-bearing.
by Addy Osmani
Sends every non-trivial decision to a fresh-context adversarial reviewer before it stands — for production, security-sensitive or irreversible work.
Persistent cross-session memory for coding agents: hooks capture each session, a local SQLite + vector store compresses it, and a mem-search skill reads it back.