AI Agent Skill Threat Landscape

Date: 2026-03-27 Audited by ClawSec

Analysis of 11411 threats detected across 1197 audited skills.

Threats by Severity

Severity Count %
CRITICAL 474 4%
HIGH 3089 27%
MEDIUM 2210 19%
LOW 5638 49%

Top 15 Threat Types

Threat Count Severity
LLM Semantic Detection 8274 LOW
Startup Failure (non-executable) 1196 LOW
Dynamic Code Evaluation 1156 HIGH
Outbound Data Transfer 363 HIGH
Hidden Command Execution 203 MEDIUM
Environment Variable Exfiltration 49 CRITICAL
Private Key Extraction 40 CRITICAL
Remote Script Execution 27 CRITICAL
Base64 Encoded Payload 27 HIGH
Systemd Service Installation 13 HIGH
Cron Job Installation 12 HIGH
Obfuscated Code 11 MEDIUM
Shell RC Modification 9 HIGH
Webhook Data Send 7 MEDIUM
SSH Key Access 6 HIGH

Threats by Verdict

Verdict Threats Avg Threats/Skill
MALICIOUS 847 11.4
SUSPICIOUS 5922 10.3
SAFE 4626 8.5

ClawSec | ClawSearch