What it does
AI coding assistants sometimes suggest packages that do not exist. Attackers can register such names and wait for someone to install them. Paste code, install commands, a package.json or a requirements.txt and the tool lists every package name it finds, compares each one with popular names, and, only if you agree, asks the public registries whether the names exist.
Step 1: find the names (offline)
- JavaScript and TypeScript:
import ... from,export ... from,import "x",require(), dynamicimport(). Scoped names such as@scope/pkgare kept and sub-paths are trimmed. Relative and absolute paths, aliases and Node built-ins (including thenode:prefix) are ignored. - Python:
import a.b,from a importandpip installlines. The standard library is ignored. Where the import name differs from the package name (for examplecv2is installed asopencv-python,PILasPillow,yamlasPyYAML), the tool says so. - Commands:
npm install,npm i,npm add,yarn add,pnpm add,bun addandpip install(alsopython -m pip,uv,poetry add). Flags and their values are skipped, and so are paths, Git sources and URLs. - Manifests:
package.json(dependencies, devDependencies, peerDependencies, optionalDependencies) andrequirements.txt.
Step 2: similarity to popular names (offline)
Each name is compared with a small built-in sample of about 300 widely used npm and PyPI names. A name is flagged when it is one or two characters away from a popular name, when only a hyphen, underscore or dot differs, or when a js or py suffix is added or dropped. The sample is not authoritative and not complete: it is only used to spot possible typos and look-alikes. Being in the sample says nothing about trust.
Step 3: registry lookup (only on request)
When you tick the consent box and press the button, the tool sends one request per package name (at most 60 names per run, 5 at a time) to registry.npmjs.org or pypi.org. For each package it shows whether it exists, the first publication date, the latest version, whether the latest version is deprecated, the number of maintainers (npm only), the repository link, install scripts (npm) and a warning when the package was first published less than 90 days ago. Errors and timeouts are shown per package.
What it cannot catch
- Trust. A package that exists can still be malicious, abandoned or a look-alike. Check the publisher, the history and the code.
- Private and internal packages. Your own workspace packages, path aliases and private registries show as not found.
- Import names. Python import names that are not in the built-in table are checked as written, and the real package name may differ.
- Dynamic names. Names built at run time or loaded from files are not seen.
- Typosquats that are not near a popular name. The sample is small.
- Registry limits. PyPI does not list maintainers in its JSON API, and very large registry records are read only in part. The tool says when something was not read.
Limits
The text can be up to 1,000,000 characters and up to 2,000 names are kept. The offline steps run in a background Web Worker. A lookup request that takes longer than 15 seconds is stopped for that package.
Example: spotting look-alike and made-up package names
AI coding assistants sometimes suggest packages that do not exist, and attackers can register such names so that an install pulls in their code. This is often called slopsquatting. The built-in example mixes real packages, typos and unknown names. The offline check, which contacts no registry, finds 12 package names and flags 3 as similar to a popular package:
lodahs: similar tolodash(two neighbouring letters swapped).expresss: similar toexpress(one character different).reqests: similar torequests(one character different).
The tool also explains import names that differ from the install name: cv2 is installed as opencv-python, PIL as Pillow and yaml as PyYAML. Names such as @scope/does-not-exist-9 get “Nothing found by the offline checks”, which does not mean they exist or are safe.
Offline first, registry only with your consent
Checking the registries is optional. The Check registries button stays disabled until you tick the consent box, because that step sends the package names, and only the names, to npm or PyPI. A name that exists is not automatically trustworthy, so also look at who publishes it, when it was created and how widely it is used.
Frequently asked questions
What does the offline check compare with?
A built-in sample of popular packages. The tool says that the sample is not authoritative, so a name that is not in it may still be fine.
Why is my own folder shown as not found?
Names that are your own folders, path aliases or workspace packages cannot be told apart from registry packages here.
Is my code uploaded?
No. Package names are extracted on your device, and nothing is sent unless you tick the consent box and run the registry check.
Privacy
The offline steps run entirely in your browser: what you paste is not sent anywhere and is not written to local storage, cookies or the URL. If you tick the consent box, only the package names shown on the page are sent, directly from your browser to the two registries.