Security Scanner
Every skill installed through Agentver is automatically scanned for security risks before it touches your system. The scanner runs 28 pattern rules across 7 categories, with zero external dependencies.
How it works
- 1. You run
agentver installoragentver audit - 2. All files are scanned against 28 pattern rules
- 3. Issues are reported with severity levels and specific line numbers
- 4. Critical/high severity issues block installation by default
Scan categories
Dangerous commands
CriticalDetects rm -rf, sudo, chmod 777, curl|bash, eval(), exec(), system() and similar destructive or arbitrary execution patterns.
Data exfiltration
CriticalDetects POST requests, fetch with body payloads, webhook.site, ngrok, requestbin, and other data exfiltration patterns.
Obfuscated code
MediumDetects long base64 strings, hex escapes, String.fromCharCode, Unicode escape sequences, and other code obfuscation techniques.
Suspicious URLs
MediumDetects pastebin links, Discord webhook URLs, URL shorteners (bit.ly, t.ly), and other suspicious remote resources.
Credential exposure
HighDetects process.env access, dynamic environment variable reading, and patterns that suggest credential harvesting.
Prompt injection
HighDetects patterns designed to override agent instructions, escape sandboxes, or manipulate agent behaviour.
Remote code execution
CriticalDetects wget, ftp, SSH tunnelling, and other patterns that enable remote code execution.
Verdicts
PASSNo issues found. Skill is safe to install.WARNMedium severity issues found. Skill is installed with warnings.BLOCKCritical or high severity issues found. Installation is blocked.Manual auditing
You can run the security scanner manually on any directory:
$ agentver audit --path ./my-skills/
scanning 12 files across 3 skills...
✓ deploy-checker PASS
✓ code-reviewer PASS
⚠ untrusted-skill WARN (2 issues)