ClawHub
Back to skill

Security audit

Subway Agent (90% OFF - Launch Special)

Security checks for vulnerabilities and agentic risk

Overview

This is an instruction-only restaurant ordering skill, but it needs review because it handles live WhatsApp orders and stores customer phone/order data in Google Sheets without clear privacy or control boundaries.

Review before installing in production. Use a dedicated WhatsApp number and least-privilege Google Sheet, add a customer privacy notice and consent language, define retention and deletion rules, require human handoff for ambiguous or high-risk cases, and verify the external ThumbGate package before running the npx setup command.

SkillSpector

Vulnerability Patterns
  • 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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Vague Triggers

Medium
Confidence
90% confidence
Finding
The trigger condition is broad enough to activate on nearly any WhatsApp message that resembles an order or inquiry, which can cause the agent to engage without clear user intent or sufficient scoping. In a customer-facing ordering context, this increases the risk of unintended data collection, accidental order handling, and inappropriate responses in conversations that should be routed elsewhere.

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The skill states that it logs every order to Google Sheets, including customer name and phone, but does not warn users that personal data will be stored in a third-party spreadsheet. This creates privacy and compliance risk because users may disclose personal information without informed consent, and operators may deploy the skill without understanding retention, access control, or legal obligations.

Ssd 3

Medium
Confidence
96% confidence
Finding
The instructions explicitly require logging every order with personal identifiers such as customer name and phone number from every conversation, creating systematic retention of personally identifiable information. In this context, Google Sheets is often loosely permissioned and not designed as a privacy-hardened datastore, so blanket logging increases the chance of unauthorized access, over-retention, and misuse of customer data.

Static analysis

No suspicious patterns detected.