OpenAI Codex Security: How to Scan a Repository and Verify AI-Generated Fixes

AI can help write code quickly. It can also introduce bugs that look perfectly reasonable in a pull request. That makes security review an increasingly important part of an AI-assisted development workflow.

OpenAI's Codex Security is an open-source CLI and TypeScript SDK for scanning repositories, investigating candidate vulnerabilities, generating patches, and checking whether fixes address the original issue.

It is also attracting attention: a GitHub trending snapshot collected on November 13, 2026 at 06:57 UTC recorded the repository second on its daily list, with 643 stars gained in that snapshot window. The number is only a snapshot; the practical question is whether the workflow fits your engineering process.

What Codex Security does

Codex Security is more than a command that prints a list of suspicious lines. Its documented workflow covers several stages:

1. Scan a whole repository, selected paths, or a Git diff.
2. Investigate candidate findings and gather source evidence.
3. Validate whether a finding is a real issue.
4. Generate a proposed patch.
5. Verify an existing fix.
6. Save reports and export findings for other tools.

It also supports threat-model work, drafting a SECURITY.md policy, classifying findings against a team's severity rubric, suggesting owners from source and Git history, and exporting results as SARIF, JSON or CSV.

The CLI and SDK are useful when you want to run scans from a terminal or integrate the workflow into automation. The Codex plugin is useful when you want an interactive investigation and patch-review loop.

Install the CLI

The current repository documentation lists these prerequisites:

• Node.js 22.x at version 22.13.0 or later, 24.x, or 26.x
• Python 3.10 or later for scans
• On Python 3.10, install tomli

Install the package globally and sign in:

npm install --global @openai/codex-security
cs login

The CLI is available as both cs and codex-security. If you do not want a global installation, use npx @openai/codex-security in place of cs.

Some requests require Trusted Access for Cyber, so check the access requirements for your account and workflow before designing a CI pipeline around it.

Run your first scan

From the root of the repository you want to review:

cs scan .

The CLI writes a Markdown report. Open the path printed as REPORT in the terminal. You can use the following commands to narrow the scope:

# Scan selected paths
cs scan . --path src --path tests

# Review committed changes against the main branch
cs scan . --diff origin/main

# Run a deeper investigation
cs scan . --mode deep

A diff scan is especially useful for a pull request workflow: it focuses attention on changes relative to a known base revision rather than repeatedly asking for a full-repository review.

Make the result useful in CI

One detail is easy to miss: according to the SDK documentation, a completed scan exits with code 0 by default even when it finds vulnerabilities.

That is reasonable for an interactive investigation, but it may not be what you want in a build pipeline. If high or critical findings should fail a job, configure the threshold explicitly:

cs scan . --fail-on-severity high

For CI, the documentation supports using OPENAI_API_KEY or CODEX_API_KEY in the scan process environment. Keep that secret in your CI secret store, scope it narrowly, and do not print it in logs.

A useful first pipeline is:

• Run a diff scan on pull requests.
• Save the generated report as a build artifact.
• Fail only at a severity threshold your team has agreed on.
• Require a human reviewer to inspect the evidence and proposed patch.
• Run a follow-up scan after the fix is merged.

Do not treat an AI-generated patch as approved just because the scanner generated it. Review the diff, tests, and security assumptions as you would any other code change.

A TypeScript SDK for custom workflows

If your team needs more control than a shell command provides, the repository also publishes a TypeScript SDK:

npm install @openai/codex-security

The package includes the Codex runtime and security plugin. The README notes that the public API may change between minor releases before version 1.0, so pin versions and test upgrades rather than assuming the API is stable.

The SDK is a better fit when you need to orchestrate scans from an internal tool, collect results, or connect the workflow to an existing engineering platform. The CLI is simpler when you only need a repeatable command in a developer terminal or CI job.

Security and operational trade-offs

This is a security tool, so the way you run it matters as much as the findings it produces.

• Run with least privilege. The project's local security model says scans run with the operating-system permissions of the process. Subprocesses can inherit environment variables.
• Protect reports. Reports and logs can contain source code, credentials, and vulnerability details. Review them before uploading or sharing.
• Keep a human in the loop. A candidate finding needs validation, and a generated patch needs code review and tests.
• Budget for model access. Some workflows require specific access and may involve model-backed investigation. Confirm eligibility and cost expectations before scaling out.
• Expect pre-1.0 change. The SDK documentation explicitly warns that minor releases may change its public API.

Codex Security should complement, not automatically replace, established controls such as CodeQL, Semgrep, dependency scanning, secret scanning, and manual security review. Those tools have different detection models and coverage. A sensible setup layers them rather than assuming one scanner catches everything.

Who should evaluate it?

It is worth testing if you maintain a TypeScript or JavaScript project, review frequent pull requests, or want a repeatable way to investigate security issues in code written with coding agents.

Start with a non-production repository. Read the report's exclusions and evidence, compare findings with your existing scanner, and measure whether the tool helps reviewers reach a confident decision faster.

Watch OpenAI's first-scan walkthrough

Sources and further reading

• Codex Security GitHub repository
• CLI and TypeScript SDK documentation
• CLI reference
• OpenAI developer documentation
• GitHub daily trending snapshot for November 13, 2026

Would you start with a full-repository scan or a pull-request diff scan in your own project?

Written by Sh Raj. Find more engineering notes on GitHub, connect on LinkedIn, and watch practical developer tutorials on StackShade.