Security audit

Smooth Browser

Security checks across malware telemetry and agentic risk

Overview

This browser automation skill appears purpose-aligned, but it gives agents broad authenticated browsing, file-upload, persistent-session, and in-page JavaScript powers without enough user control or warning.

Install only if you intentionally want an agent to control a browser through Smooth. Use separate browser profiles for this skill, avoid sensitive accounts unless necessary, confirm every file upload, avoid secrets or regulated documents, restrict JavaScript execution to trusted pages, and clear stored profiles/session mappings when the task is done.

SkillSpector

By NVIDIA

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (5)

Vague Triggers

High

Confidence: 95% confidence
Finding: The trigger description is extremely broad and can cause this skill to activate for many generic web-related requests, including requests involving credentials, payments, or sensitive browsing tasks. Because the skill enables authenticated browsing, scraping, file upload, and automation, overbroad routing increases the chance that sensitive tasks are silently delegated to an external browser service without sufficient user awareness or consent.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: The workflow discusses profiles, sessions, cookies, authentication, and persistent state, but does not prominently warn that these artifacts may contain sensitive account data and may persist across tasks. This can lead users or downstream agents to reuse logged-in profiles or session state in ways that exceed the original user intent, exposing personal or organizational accounts.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: The file-upload workflow instructs the agent to upload local files to the external Smooth service for browser-session use, but it does not clearly warn that documents are leaving the local environment. Users may unintentionally expose confidential files, regulated data, or proprietary material if the agent follows these instructions automatically.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: The skill documents arbitrary JavaScript execution in the browser context without cautioning that injected code can read page contents, manipulate forms, exfiltrate visible data through subsequent actions, or alter authenticated application state. In an agentic setting, this materially increases risk because prompts or intermediate data could lead to unsafe script execution against sensitive sites.

Ssd 3

Medium

Confidence: 94% confidence
Finding: The instruction to save which profiles authenticate to which services for later reuse encourages retention of durable account-access mappings beyond the immediate task. That creates a path for cross-task privilege carryover, where later requests may gain access to previously authenticated services without fresh user intent or awareness.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.