verify-evidence-loop
Warn
Audited by Socket on Apr 19, 2026
1 alert found:
AnomalyAnomalySKILL.md
LOWAnomalyLOW
SKILL.md
Mostly coherent research skill with good HITL and injection-awareness, but its permission footprint is broader than necessary: it combines untrusted web content ingestion with Write and especially Bash access. No credential harvesting, malicious data flow, or suspicious installer behavior is present. Overall this is better classified as suspicious/medium-risk due to indirect prompt-injection exposure and unnecessary execution capability, not malware.
Confidence: 89%Severity: 61%
Audit Metadata