Secure Skills
for your agents
Verify AI agent skills before installation. Multi-stage analysis combining YARA pattern detection, LLM semantic review, and sandboxed execution monitoring to catch threats before they reach your agents.
terminal
$ clawsec install clawhub.ai/skill-name
Verifying skill integrity...
[STAGE_1]Pattern Detection
└──
YARA rules scan(<1s)
[STAGE_2]Intelligence Analysis
└──
LLM semantic review(5-10s)
[STAGE_3]False Positive Filter
└──
Meta-analysis(5-10s)
[STAGE_4]Live Testing
└──
Sandbox execution(coming soon)
→Trust Score:95/100
→Risk Level:LOW
ANALYSIS_COMPLETE
Installing skill...
✓Installation complete
$
Recent Scans
Latest verification results
LOADING SCANS...
Analysis Pipeline
Three-stage verification system
STAGE 1 // STATIC ANALYSIS
YARA Pattern Detection
Custom YARA rules scan skill source code for known malicious patterns, obfuscation techniques, and suspicious code structures. Results are instant and require no external API calls.
- +Prompt injection & coercion detection
- +Command & code execution patterns
- +Credential harvesting signatures
- +Unicode steganography identification
- +Privilege escalation attempts
Detection Rules13 RULES
prompt_injection_genericCRITICAL
coercive_injection_genericCRITICAL
command_injection_genericCRITICAL
code_execution_genericCRITICAL
credential_harvesting_genericCRITICAL
sql_injection_genericCRITICAL
tool_chaining_abuse_genericHIGH
system_manipulation_genericHIGH
prompt_injection_unicode_steganographyHIGH
autonomy_abuse_genericMEDIUM
capability_inflation_genericMEDIUM
data_exfiltration_genericHIGH
privilege_escalation_genericCRITICAL