AI Agent Skill Threat Landscape
| Date: 2026-03-29 |
Audited by ClawSec |
Analysis of 19272 threats detected across 1948 audited skills.
Threats by Severity
| Severity |
Count |
% |
| CRITICAL |
956 |
5% |
| HIGH |
5362 |
28% |
| MEDIUM |
3802 |
20% |
| LOW |
9152 |
47% |
Top 15 Threat Types
| Threat |
Count |
Severity |
| LLM Semantic Detection |
14002 |
LOW |
| Startup Failure (non-executable) |
1934 |
LOW |
| Dynamic Code Evaluation |
1871 |
HIGH |
| Outbound Data Transfer |
695 |
HIGH |
| Hidden Command Execution |
312 |
MEDIUM |
| Environment Variable Exfiltration |
97 |
CRITICAL |
| Private Key Extraction |
95 |
CRITICAL |
| Base64 Encoded Payload |
54 |
HIGH |
| Remote Script Execution |
47 |
CRITICAL |
| Shell RC Modification |
23 |
HIGH |
| Cron Job Installation |
22 |
HIGH |
| Systemd Service Installation |
17 |
HIGH |
| Obfuscated Code |
17 |
MEDIUM |
| Cryptocurrency Wallet Access |
17 |
CRITICAL |
| Webhook Data Send |
15 |
MEDIUM |
Threats by Verdict
| Verdict |
Threats |
Avg Threats/Skill |
| MALICIOUS |
1917 |
12.1 |
| SUSPICIOUS |
9887 |
10.6 |
| SAFE |
7452 |
8.7 |
ClawSec | ClawSearch