Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Real Packages Only

    1

    AI coding assistants invent package names that sound right, or pick a near-miss of a real one.

    $9

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Real Packages Only

    Real Packages Only

    Example session with this skill installed

    Before I install anything: check that every dependency in this project is a real package and safe to install.

    • Read your context and instructions
    • Compiled the real packages only

    A real, unedited answer from a test run (Claude Sonnet, Claude Code, 2026-11-12). The sample project petsitter-api had nine npm dependencies written by an AI assistant. Version numbers are what the npm registry returned on the day of the run.


    ❌ DO NOT INSTALL YET

    I checked all 9 dependencies in package.json against the npm registry. I only read registry metadata and ran no installs. 4 of the 9 are a problem.

    PackageVerdictEvidence
    express✅ RealPublished since 2010, official expressjs/express repo.
    zod✅ RealPublished since 2020, 1000+ versions, official repo.
    dotenv✅ RealPublished since 2013, official repo.
    is-odd✅ RealPublished since 2015, by jonschlinkert. It's silly but legitimate.
    left-pad🟡 Real but deprecatedMarked "use String.prototype.padStart()". Last release was in 2024. It's safe to keep, but you don't need it.
    express-session-redis-store-pro❌ Not realThe registry returns 404.
    stripe-webhook-verify-easy❌ Not realThe registry returns 404.
    lodahs🟡 Typosquat, placeholderIt's a misspelling of lodash. The only version is 0.0.1-security, a "security holding package" from npm's security-holder repo.
    faker🟡 Wrong packageMaintainer-sabotaged release line. See below.

    Problems

    express-session-redis-store-pro (❌)

    • This name looks AI-invented, and nothing is published under it.
    • Someone could register it later and put malware behind it. That's the slopsquatting risk.
    • The code needs a Redis session store. The real packages are express-session and connect-redis.
      • express-session is at 1.19.0 and was published in 2014. The repo is expressjs/session.
      • connect-redis is at 10.0.0 and was published in 2010. The repo is tj/connect-redis.
    • Fix: npm i express-session connect-redis, plus a redis or ioredis client.

    stripe-webhook-verify-easy (❌)

    • This name is also not published.
    • Stripe's official stripe package (23.1.0, from stripe/stripe-node) verifies webhook signatures with stripe.webhooks.constructEvent.
    • Fix: npm i stripe.

    lodahs (🟡)

    • This is a typo of lodash.
    • It's an empty placeholder today, but it's the kind of name attackers register.
    • Fix: replace it with lodash (real, 4.18.1, from lodash/lodash). Then change require('lodahs') to require('lodash') in src/server.js.

    faker (🟡)

    • The original faker is the Marak/Faker.js package. Its maintainer sabotaged the 6.6.6 release in January 2022, and the project was abandoned.
    • This is from my own knowledge. The registry data I pulled doesn't mention it.
    • The registry confirms the version pin 6.6.6 is the latest, with no deprecation notice.
    • The community fork is @faker-js/faker (real, 10.6.0, from faker-js/faker).
    • It's also a dev-data tool, so it shouldn't normally be a production dependency.
    • Fix: npm i -D @faker-js/faker, then require('@faker-js/faker').

    Other things I noticed

    • Your version ranges are old (express ^4, zod ^3, dotenv ^16). Newer majors exist, but I didn't check compatibility, since that's outside this check.
    • I didn't run a CVE or vulnerability scan. That's a separate step if you want it.

    I can apply these fixes to package.json and src/server.js if you want. I wouldn't run the install yet.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    AI coding assistants invent package names that sound right, or pick a near-miss of a real one. Research on code-generating models found roughly 1 in 5 suggested packages didn't exist, and the same invented names come up again and again. Attackers register those names with malware inside ("slopsquatting"), so the next npm install or pip install runs their code on your machine.

    Worse are the packages that install fine but are wrong: pip install jwt gets you a different library than the PyJWT your code calls, dotenv is a deprecated stub, and on npm fs, http and axios-http are empty security placeholders. Nothing errors until much later.

    Real Packages Only makes your assistant check first. Before anything is installed, every package is looked up on the official registry and rated:

    • ❌ Not real: nothing published under that name. Never installed.
    • 🟡 Check first: exists, but it's a look-alike of a popular package, a security placeholder, a deprecated stub, brand new with no history, or the wrong package for the import (cv2 → opencv-python, PIL → Pillow, sklearn → scikit-learn)
    • ✅ Real: the right package, normal history
    • ⚪ Not checked / Private: no registry access, or your company's own package (never guessed)

    For every problem you get the right package (also verified), the corrected install command, and a copy-paste prompt to fix it.

    Two modes

    • Guard mode: kicks in whenever your assistant is about to run npm install, pip install, cargo add, go get and friends, or add a dependency. Real packages go ahead; bad ones are stopped with a suggestion.
    • Audit mode: "Are all the packages in this project real?", "Check my requirements.txt", "What do I pip install to run this script?"

    Built to be safe

    • Never installs to check. Installing runs a package's code, so it only reads registry metadata, and it knows which commands secretly download and run code (npx, uvx, pnpm dlx)
    • The registry decides, not the AI's memory. No network? It says "not checked" instead of guessing
    • No false alarms on odd-but-real packages (is-odd, left-pad, bs4)
    • 8 registries: npm, PyPI, crates.io, Go modules, RubyGems, Packagist, NuGet, Maven Central
    • No scripts, no setup, nothing to install

    Tested on 9 sample projects (Node, Python, Rust, with planted fake, look-alike, placeholder and wrong-import packages) in blind Claude Code runs with Sonnet and Haiku: every fake package caught, zero false alarms on real packages, and on plain "install the dependencies" requests it stopped the silent traps that the same models installed without the skill.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Security.

    Does not scan for vulnerabilities (CVEs) or malware in real packages; keep using npm audit, pip-audit, Dependabot or Socket for that. Can't verify anything offline, and says so instead of guessing. Private packages are marked "Private, not checked" unless you ask it to check your own registry. Smaller, faster models occasionally skip the skill on a plain "install the dependencies" request; the README gives a one-line rule for CLAUDE.md / AGENTS.md that makes it run every time. Tested end to end in Claude Code with Sonnet and Haiku; other agents not yet tested end to end.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 3 days ago

    Accounts
    Crates
    Software
    GitPython

    Permissions

    Terminal / Shell
    Network Access
    Read Files
    Write Files

    Allowed Hosts

    registry.npmjs.org
    crates.io

    Terminal or shell is optional: read-only registry lookups with curl or `npm view`, and `git diff` for recent changes. Files: reads manifests and code; writes `package-check-report.md` only for long lists or when asked; edits manifests only if the user asks for fixes. Network: public package registry APIs only (registry.npmjs.org, pypi.org, crates.io, proxy.golang.org, rubygems.org, packagist.org, api.nuget.org, search.maven.org). No accounts or services to connect.

    Listed3 days ago

    What's inside

    Frequently Asked Questions

    No. It checks that packages exist and are the ones you meant, which is where AI assistants go wrong. Keep using `npm audit`, `pip-audit`, Dependabot or Socket for vulnerabilities.

    A few seconds per new package: one registry lookup each, batched. Well-known packages in a big project get a quick existence check only.

    They're marked "Private, not checked" and left alone unless you ask it to check your own registry.

    It can't verify anything offline, and it says so instead of guessing.

    With strong models, yes in our tests. Smaller, faster models occasionally skip a skill on a plain "install the dependencies" request, so the README gives a one-line rule to add to CLAUDE.md / AGENTS.md that makes it run every time.

    Only if you ask it to fix the names. It never installs a package that didn't check out.

    New in Security