Security audit

DeepSeek Web Query

Security checks across malware telemetry and agentic risk

Overview

This skill appears intended to query DeepSeek, but it can send broad user prompts through a logged-in browser session, read clipboard content, and change local MCP tooling without enough guardrails.

Install only if you are comfortable with selected prompts being sent to DeepSeek through an authenticated browser. Use it with explicit DeepSeek requests, avoid secrets or proprietary content, keep sensitive clipboard data clear, and review the MCP recovery steps before allowing it to modify mcporter configuration or run an unpinned npx package.

SkillSpector

By NVIDIA

Vulnerability Patterns

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

Findings (5)

Missing User Warnings

Medium

Confidence: 87% confidence
Finding: The document explicitly instructs the skill to click a copy button and read answer content from the system clipboard, but it does not warn that clipboard access can expose unrelated sensitive data if clipboard contents are read unexpectedly or permissions are broader than assumed. In this skill context, browser automation plus session reuse increases risk because the agent may operate in a logged-in browser environment and interact with user data without clear privacy boundaries.

Vague Triggers

Medium

Confidence: 84% confidence
Finding: The trigger conditions are broad enough that ordinary requests like 'search一下' or general knowledge questions may invoke this skill automatically and forward user content to DeepSeek. In context, that increases the chance of accidental disclosure of sensitive prompts or use of a third-party web service when the user did not intend external transmission.

Vague Triggers

Medium

Confidence: 85% confidence
Finding: The 'When to Use' section lacks clear boundaries, so the skill may be selected for many generic requests unrelated to its intended purpose. Because the skill sends queries to a third-party website and maintains browser state, accidental activation creates privacy and data handling risks beyond a normal local-only assistant action.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The skill forwards user queries to a third-party website through automated browser interaction but does not prominently warn about that data transfer. This is dangerous because users may provide confidential prompts, source code, or personal data under the assumption that processing remains within the primary assistant environment.

Missing User Warnings

Low

Confidence: 82% confidence
Finding: The instruction to keep the browser session open for reuse increases persistence of authenticated state, conversation history, and potentially sensitive page data, but the privacy implications are not clearly disclosed. This can enable later unintended access from subsequent tasks or users sharing the same environment.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.