ClawHub
Back to skill

Security audit

Coupang

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed Coupang connector skill with business-account write capability that is scoped to the oo CLI and requires confirmation for state-changing actions.

Install this only if you intend to let Codex use your OOMOL-connected Coupang account. Review write payloads carefully before approving price or inventory changes, and consider manually verifying the oo CLI installer instead of piping a remote script directly to a shell.

SkillSpector

Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • 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 (3)

Intent-Code Divergence

Medium
Confidence
95% confidence
Finding
The skill labels `get_order` as `[write]` even though its description says it only retrieves order details. Mislabeling action semantics can cause agents or users to apply the wrong safety policy, either blocking legitimate reads or, more importantly, normalizing inaccurate trust boundaries in a skill that also performs real write operations. In a commerce integration, incorrect action classification increases the chance of unsafe automation decisions and operator confusion.

Vague Triggers

Medium
Confidence
90% confidence
Finding
The instruction to use this skill for ANY Coupang request creates an overly broad delegation boundary, encouraging the agent to route all Coupang-related tasks through a shell-capable wrapper without narrowing by operation type or risk. Because the skill supports both reads and writes, this broad scope can cause unnecessary invocation of a powerful integration and increases the blast radius of prompt mistakes, ambiguous requests, or malicious user phrasing.

Missing User Warnings

Medium
Confidence
97% confidence
Finding
The first-time setup recommends `curl ... | bash` and the PowerShell equivalent, which execute remote code directly without verification or a warning. If the install endpoint, network path, or hosting environment is compromised, an attacker could gain arbitrary code execution on the user's machine. In a skill document, this is especially dangerous because it may be presented as a routine troubleshooting step and followed without scrutiny.

Static analysis

No suspicious patterns detected.