More screenshots
Works with the AI Agents you already use
Fake Package Blocker: Stop AI-Hallucinated Dependencies
Turn every pip/npm/cargo install your agent runs into a checked one: blocks typosquats and hallucinated package names before they reach your machine.
$15
About this skill
Your agent types pip install huggingface-cli or adds crossenv to package.json. It looks right to the agent, because the agent's own guess produced it. Attackers register exactly those names.
Replayed against 629 real malicious typosquat uploads, the offline check blocks 44% and flags 72.5% before anything installs. It wrongly blocked 1 of 2,193 real dependencies in 10 public repos that were kept out of tuning.
Code models invent package names: 19.7% of 2.23M samples from 16 models named at least one package that does not exist (USENIX Security 2025). Phantom Dependency Guard gives the agent the data it lacks: which names are popular, how close a new name is to one of them, and what the registry says.
What the tool does
- Reads install commands (pip, uv, poetry, pdm, pipx, npm, pnpm, yarn, bun, npx, cargo, go, gem, composer) and the dependencies added to requirements.txt, pyproject.toml, package.json, Cargo.toml, go.mod, Gemfile and composer.json.
- Offline, it checks each name against the 15,000 most-used PyPI and npm packages plus top crates, gems, Composer and Go modules. It catches keyboard slips, look-alike characters (
jeIlyfish), separator tricks (crossenv), fake npm scopes, Go owner copies (boltdb-go/bolt) and affix confusions (aws-cdkforaws-cdk-lib), and names the real package. - With
--online(opt-in, 4 s timeout) it also asks the registry: missing package, npm security placeholder, a first release under 30 days old, almost no downloads, install scripts. - A Claude Code hook blocks the install before it runs and asks you on warnings, with a visible allow list and off switch. Plus
scan,diff --base mainand a GitHub Actions file.
Benchmark (all commands and commit SHAs are in the docs)
| Set | Offline flagged | Online blocked |
|---|---|---|
| 629 real typosquat uploads (npm 2017, PyPI 2017/2019/2024, Go, crates) | 72.5% (44% blocked) | 99.5% |
| 25 LLM-hallucinated names from published research | 11 | 23 (+1 warned) |
| 2,193 real dependencies, 10 held-out repos (grafana, mlflow, zed...) | 1 false block | 1 false block |
Most of the online blocks are packages the registries have already removed. A fresh squat depends on the name signals plus the age and download checks. On Home Assistant's 1,178 tiny device libraries it raised 8 false blocks, so short niche names need the allow list.
What it does not do: it does not scan package code for malware, check transitive dependencies or spot a correctly named but compromised package. Offline, it cannot tell that a brand-new invention far from every popular name is fake.
Works in Claude Code, Cursor, Codex CLI, Copilot and any SKILL.md agent.
Privacy: offline by default, no telemetry. --online sends only package names to the public registries.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse All Claude Skills.
How to use
- 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
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 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 27 days ago
- Accounts
Crates
Packagist
Ecosyste
Pypistats
Golang
Rubygems
- Software
Pythonpkg
Node.js
Git
Permissions
Allowed Hosts