neuralos

Warn

Audited by Socket on Jul 31, 2026

4 alerts found:

SecurityAnomalyx3
SecurityMEDIUM
SKILL.md

SUSPICIOUS: the skill is broadly consistent with a headless RTerm/neuralOS operations backend, so it is not fundamentally deceptive, but its footprint is extremely expansive and high-impact. It combines autonomous remote execution, credentialed integrations, external content ingestion, background services, and real-world actions, making it a high-risk skill even without clear evidence of malware.

Confidence: 86%Severity: 84%
AnomalyLOW
examples/schedule-weekly-av.mjs

Overall, the code functions as an administrative remote automation client: it connects to a configurable WebSocket gateway, reads existing scheduler settings, and then pushes a new scheduled task whose command is a PowerShell AV signature update (Update-MpSignature). The main security concern is the high-impact remote configuration write that results in later PowerShell execution by the backend/scheduler; misuse depends on gateway trust and server-side authorization. No clear evidence of stealth, obfuscation, or direct data theft is present in this snippet, but the capability to schedule PowerShell remotely warrants careful trust boundary review.

Confidence: 62%Severity: 60%
AnomalyLOW
service/install-windows-service.ps1

This PowerShell module is an installer/persistence utility rather than an obvious in-file malware payload: it does not contain credential theft or obfuscated logic. However, it introduces meaningful supply-chain and operational risk by performing an unpinned npm global install at runtime and then creating a persistent, aggressively auto-restarting scheduled task that launches the installed backend. It also configures the service to listen on 0.0.0.0:17888, increasing exposure if authentication/TLS is not robust in the backend.

Confidence: 62%Severity: 63%
AnomalyLOW
examples/fleet-health-gate.mjs

No overt local malware behavior is evident in this file (no exec/eval, no filesystem writes, no direct credential theft or network exfiltration). However, it delegates operational actions to a remote agent via free-form, command-like userInput and determines CI pass/fail solely from remote UI text, creating meaningful supply-chain/trust and integrity risk if the WebSocket endpoint or remote agent is not fully trusted. Additionally, the absolute-path fallback when requiring 'ws' is an unusual environment-specific anomaly that warrants review for packaging correctness and unintended behavior.

Confidence: 62%Severity: 50%
Audit Metadata
Analyzed At
Jul 31, 2026, 07:23 PM
Package URL
pkg:socket/skills-sh/DrOlu%2Fagent-skills%2Fneuralos%2F@706297395ce66056a980e3d894f8855d569b442be84f4bea4823a5f1414910cf
Security Audit — socket — neuralos