Security audit

Even G2 Bridge

Security checks across malware telemetry and agentic risk

Overview

This is a coherent G2-to-OpenClaw bridge, but it creates a high-authority voice-controlled internet path to the main agent and can automatically forward raw prompts and results to third-party services.

Install only if you are comfortable exposing your OpenClaw agent through a Cloudflare Worker controlled by G2 voice commands. Use a strong G2_TOKEN, verify unauthenticated POST requests return 401, prefer a scoped Gateway token or separate low-privilege agent, keep confirmations on mutating tools, and enable Telegram/OpenAI/Anthropic only for content you are willing to send to those services.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep

Findings (5)

Missing User Warnings

Medium

Confidence: 87% confidence
Finding: The skill states that long-task results and generated images may be delivered through Telegram, but it does not prominently warn users that their prompts, outputs, or derived content may be transmitted to a third-party service. This creates a privacy and data-handling risk, especially if users send sensitive voice content, agent outputs, or generated material through the bridge without realizing it leaves the primary system boundary.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: User prompts are sent to Anthropic in the fallback path and to OpenAI during image generation, but the code provides no explicit consent flow or strong user-facing disclosure that third-party providers will receive the content. In a voice-assistant context, users may reasonably expect processing by the primary gateway only, so silent routing to additional external AI providers creates a privacy and data-governance risk.

Missing User Warnings

High

Confidence: 97% confidence
Finding: Long-task handling forwards the full user task and resulting model output to Telegram automatically, which can expose sensitive prompts, generated code, research, or operational details to an external messaging platform. Because this happens in the background based on pattern matching rather than explicit confirmation for each request, users can unintentionally leak private or regulated data.

Ssd 3

Medium

Confidence: 96% confidence
Finding: The code automatically creates plain-language Telegram summaries containing the original task text and the generated result, which is a direct data exfiltration path to a third-party channel. In this skill's context, long tasks include coding, research, deployment, and translation requests, so the exported content may contain proprietary source code, credentials accidentally pasted by users, or sensitive business information.

Ssd 3

Low

Confidence: 89% confidence
Finding: Image-generation prompts are copied into Telegram photo captions, exposing whatever the user requested to an external messaging service. Although lower impact than full result forwarding, prompts can still contain personal, confidential, or sensitive information and are sent without an explicit consent checkpoint.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.