ClawHub
Back to skill

Security audit

E-commerce Gmail Customer Service

Security checks across malware telemetry and agentic risk

Overview

This is a powerful Gmail customer-service automation skill, but its access, memory, sending, scheduling, and setup controls are clearly disclosed and purpose-aligned.

Install only for a dedicated support Gmail account, keep auto-send off until draft-only tests pass, review the category permission file before enabling any automatic sends, and periodically inspect or clear long-term memory. Store OAuth and merchant credentials only in approved secret stores, not in chat, Gmail, config files, or the skill directory.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (2)

Vague Triggers

Medium
Confidence
92% confidence
Finding
The configured human-handoff phrase is a common natural-language expression that a customer could easily include in a normal reply, intentionally or accidentally. In this skill, that phrase appears to control escalation to manual handling, so ambiguity can let customers manipulate routing, disrupt automation, or cause unnecessary human-review workload and inconsistent case handling.

Session Persistence

Medium
Category
Rogue Agent
Content
---
name: ecommerce-gmail-customer-service
description: "Draft-first e-commerce Gmail support: triage customer threads, verify order and policy context, and create auditable drafts. Owners independently control ongoing draft-edit learning, existing long-term memory use, and category-based automatic sending."
version: 1.2.8
metadata:
  openclaw:
Confidence
78% confidence
Finding
create auditable drafts. Owners independently control ongoing draft-edit learning, existing long-term memory use, and category-based automatic sending." version: 1.2.8 metadata: openclaw: requir

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.