A Claude skill that audits your project's dependencies against OSV.dev — Google's open vulnerability database — and flags compromised, malicious, or vulnerable package versions before they reach production.
- Walks your project directory and finds all dependency files
- Parses them into
(package, version, ecosystem)triples — lockfiles preferred over manifests - Batch-queries OSV.dev for known issues
- Reports findings in two buckets:
- Malicious / compromised — remove or roll back (supply-chain attacks, worms, exfiltrators)
- Vulnerable — upgrade to the fixed version (regular CVEs)
| Ecosystem | Files parsed |
|---|---|
| npm | package-lock.json, yarn.lock, pnpm-lock.yaml, package.json |
| PyPI | poetry.lock, Pipfile.lock, uv.lock, requirements.txt, pyproject.toml |
| Cargo | Cargo.lock |
| Go | go.sum |
| RubyGems | Gemfile.lock |
| Composer | composer.lock |
| Maven | pom.xml |
| NuGet | *.csproj, packages.lock.json |
Install compromised-package-checker.skill through Claude's skill settings, then ask:
- "Check my project for compromised packages"
- "Audit my dependencies"
- "Am I affected by Shai-Hulud?"
- "Scan my lockfile for malicious packages"
# Full audit
python3 scripts/check.py /path/to/your/project
# Malicious packages only
python3 scripts/check.py /path/to/your/project --malicious-only
# JSON output (for CI pipelines)
python3 scripts/check.py /path/to/your/project --jsonNo external dependencies — pure Python stdlib. PyYAML is optional (improves YAML parsing).
- name: Audit dependencies
run: |
python3 scripts/check.py . --json > audit.json
python3 -c "
import json, sys
d = json.load(open('audit.json'))
mal = d['totals']['malicious']
crit = sum(1 for f in d['findings'] if f['severity'] == 'CRITICAL')
sys.exit(1 if mal or crit else 0)
"| Code | Meaning |
|---|---|
0 |
Clean — no findings |
1 |
Findings present |
2 |
Error (no lockfiles found, parse failure) |
All vulnerability data comes from OSV.dev — a free, read-only, key-free API that aggregates:
- GitHub Advisory Database (GHSA)
- PyPA, RustSec, Go vuln DB
- npm audit feed
- OSSF Malicious Packages Repository (
MAL-*IDs)
A clean scan means "nothing known today." Zero-days and fresh compromises may take hours to appear in OSV.
- Shai-Hulud — self-replicating npm worm (Sept 2025, 500+ packages)
- s1ngularity — credential exfiltrator across PyPI and npm
- chalk/debug compromise — phishing-driven maintainer takeover
- Any package with an active
MAL-*OSV advisory
MIT