Security audit

Global Business AI Assistant

Security checks across malware telemetry and agentic risk

Overview

This is a coherent WLdPass business-integration skill with real privacy and credential-handling considerations, but the sensitive behavior is mostly disclosed and aligned with its purpose.

Install only if you trust WLdPass with the business data you submit through this skill. Use a dedicated or revocable WLdPass token, protect ~/.openclaw/openclaw.json, confirm before letting an agent send messages, post demands, create public/community content, or enable auto-replies, and register webhooks only to endpoints you control because they may receive future message, match, contact, watch, and digest events.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
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)

Intent-Code Divergence

Medium

Confidence: 87% confidence
Finding: The script description claims setup steps remain reviewable locally, but the helper performs live outbound API calls to verify tokens and optionally register a webhook. This is not code execution or stealth exfiltration by itself, but it is a security-relevant transparency issue because users are not clearly warned in the header/usage text that the script will contact remote services and modify remote account state.

Missing User Warnings

Medium

Confidence: 86% confidence
Finding: The skill explicitly exposes features for unlocking contact details and retrieving public contacts, but it does not warn the user that using these actions may disclose or aggregate personal/business contact information. In an agent-driven setting, this increases the risk of automated privacy-invasive collection, outreach, or downstream misuse without meaningful user awareness or consent checks.

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: The webhook and scheduled report features send potentially sensitive business data to external URLs or delivery channels, but the skill lacks explicit warnings that this constitutes external data transmission. In an autonomous-agent context, users may unknowingly route messages, matches, contacts, or digest data to third-party systems, increasing confidentiality and data governance risk.

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: The helper stores the provided API token in ~/.openclaw/openclaw.json without clearly warning the user beforehand that credentials will be written to disk. Persisting secrets in config files can increase exposure through local compromise, backups, accidental sharing, or overly broad file permissions, especially for agent ecosystems that may load and propagate environment values.

Missing User Warnings

Medium

Confidence: 84% confidence
Finding: The webhook registration code performs an authenticated outbound POST to a remote endpoint, changing remote account configuration, but this behavior is only implied by the --webhook flag and not clearly disclosed as a security-sensitive network action in the script description. While expected for setup tooling, lack of explicit notice reduces informed consent and can surprise users into exposing operational data flows to a URL they specify.

VirusTotal

67/67 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.