Codex Token Guard

Local security checks for Codex and OpenAI developer environments on Windows.

Codex Token Guard looks for suspicious token-like values, risky local AI coding settings, and Codex/OpenAI-looking global npm packages that deserve manual verification. It is designed for developers who want a fast local safety check after installing Codex-related tooling, plugins, scripts, or packages.

Why this exists

AI coding tools create a new local attack surface:

• copied install commands from social posts,
• npm packages with convincing names,
• local config files that may contain credentials,
• plugin or MCP scripts that touch browser, shell, or account state.

This project does not claim to prove that a machine is safe. It gives you a quick local report so you can find obvious risks before sharing logs, publishing a repo, or continuing to use a suspicious setup.

Safety model

Codex Token Guard is intentionally conservative:

• It does not upload files.
• It does not print full token values.
• It redacts secret-like matches in reports.
• It scans a small default scope unless you explicitly pass more paths.
• It does not delete, revoke, uninstall, or modify anything.

Quick start

Run a default local check:

powershell -ExecutionPolicy Bypass -File .\scripts\codex-token-guard.ps1

Check global npm packages with Codex/OpenAI-looking names:

powershell -ExecutionPolicy Bypass -File .\scripts\codex-token-guard.ps1 -CheckNpm

Scan a specific project or folder:

powershell -ExecutionPolicy Bypass -File .\scripts\codex-token-guard.ps1 -Path C:\path\to\project

Generate a Markdown report:

powershell -ExecutionPolicy Bypass -File .\scripts\codex-token-guard.ps1 -Path C:\path\to\project -Format Markdown

What it checks

Current checks include:

• OpenAI API-key-looking strings such as sk-...
• session-token-looking strings such as sess-...
• bearer-token-looking strings
• inline variables such as OPENAI_API_KEY=..., AUTH_TOKEN=..., CLIENT_SECRET=...
• global npm packages with names containing codex, openai, chatgpt, or gpt that are not on a tiny official allowlist

The npm package check is not a malware verdict. It is a triage signal: verify the publisher, source repository, install command, package history, and security reports before trusting the package.

Exit codes

• 0: no high or medium findings
• 2: at least one high or medium finding

Recommended response to findings

If a real token was exposed:

1. Revoke or rotate it first.
2. Remove it from local files and git history if needed.
3. Check whether the token was committed, logged, uploaded, or pasted into a public issue.
4. Move future secrets to environment variables, ignored local files, or a secret manager.

If a suspicious package was found:

1. Verify the package name and publisher.
2. Check the source repository and install command.
3. Search for recent security reports.
4. Remove it if you cannot explain why it is installed.

Limitations

• Pattern matching can produce false positives and false negatives.
• Redaction reduces leakage risk but does not make reports automatically safe to publish.
• This tool cannot confirm whether a token is valid.
• This tool cannot confirm whether an npm package is malicious.
• Large files and common dependency/build folders are skipped by default.

Related project

For Windows Codex plugin troubleshooting, see codex-windows-plugin-doctor.

License

MIT