ClawHub
Back to skill

Security audit

Taobao Query

Security checks for vulnerabilities and agentic risk

Overview

This Taobao shopping skill is disclosed, but it gives broad account-linked powers like cart changes, order/history reading, and seller chat without enough scoping or confirmation.

Review this skill before installing. Use it only with a trusted local or TLS-protected MCP server, and require explicit confirmation before it reads orders/history, adds items to a cart, navigates account pages, or sends seller messages. Do not treat it as a read-only price checker.

A.I.G

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:22
Finding

Remote MCP Connections Permit Plaintext Transmission of Sensitive Session Data

Content
View full analysis
` 3. **Call tools** with session ID in header: `mcp-session-id: ` ``` ### Technical Analysis The default loopback URL has limited network exposure, but the Skill explicitly recommends configuring non-loopback servers such as `http://192.168.100.20:3654/mcp` without TLS. HTTP provides neither transport confidentiality nor server authenticity. When a remote or LAN endpoint is used, MCP initialization requests, subsequent tool calls, returned commerce information, and the `mcp-session-id` header can cross the network in plaintext. An attacker with a suitable network position could inspect or modify this traffic. If the MCP server treats possession of the session ide ...[truncated 1912 chars]
Remediation
View remediation

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (8)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The description expands the skill from product-price lookup into orders, cart, and seller chat without clearly constraining when those account-linked operations may be used. This ambiguity can lead an agent to invoke higher-risk capabilities under the umbrella of a benign query skill, exposing account data or sending messages unexpectedly.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The description expands the skill from product-price lookup into orders, cart, and seller chat without clearly constraining when those account-linked operations may be used. This ambiguity can lead an agent to invoke higher-risk capabilities under the umbrella of a benign query skill, exposing account data or sending messages unexpectedly.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill marks 'add items to cart' as permitted and 'safe,' but adding to cart changes account state and can influence later purchases, recommendations, or promotions. Without an explicit user-facing warning and confirmation requirement, the agent may perform an unintended modification to the user's shopping account.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
84% confidence
Finding

The skill explicitly sends data to an MCP HTTP endpoint and even encourages configuring a non-local server IP, which creates a channel for transmitting user queries and account-derived data outside the local agent boundary. In this commerce context, transmitted content can include search terms, cart/order details, or chat actions, so the external communication path is security-relevant.

Content

Scanner excerpt · SKILL.md (reported line 115)May include surrounding context.

  1. Initialize connection to get session ID:
bash
curl -X POST ${TAOBAO_MCP_URL:-http://127.0.0.1:3654/mcp} \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

Documenting browsing-history access without a privacy warning normalizes retrieval of potentially sensitive behavioral data such as searched products, viewed shops, and purchase intent. In an account-linked commerce context, this can expose private preferences or household information beyond what is needed for a typical price-query task.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The output templates are written entirely in Chinese and presented as the required response formats, which implies the skill will answer in Chinese regardless of user preference. There is no documented opt-in or language-selection behavior to let users choose another language.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
78% confidence
Finding

The health-check example performs another outbound MCP call, which is an external transmission path, but by itself it is lower risk than the main operational calls because it is limited to session and tab-state checking. It is still security-relevant because it reinforces routine transmission to a potentially remote MCP endpoint and may disclose page/context metadata.

Content

Scanner excerpt · SKILL.md (reported line 302)May include surrounding context.

Before major operations (e.g., viewing cart with many items), optionally verify connection:

bash
# Quick health check - try to get current tab
curl -X POST ${TAOBAO_MCP_URL:-http://127.0.0.1:3654/mcp} \
  -H "mcp-session-id: <session-id>" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"get_current_tab"}}'

Scope Creep

Low
Category
Excessive Agency
Confidence
90% confidence
Finding

The phrase 'can also perform other Taobao operations' broadens the effective capability surface beyond the named purpose of querying prices, which is classic scope creep. Even with a payment prohibition, expanding into orders, carts, and chat increases the chance of accidental account access or action under a skill users may perceive as read-only.

Content

Scanner excerpt · SKILL.md (reported line 3)May include surrounding context.

md
---
name: taobao-query
description: Query Taobao product prices and information through MCP (Model Context Protocol). Use when the user asks about product prices, searches for items on Taobao, compares prices, or wants to find the best deals on specific products. Triggers on phrases like "淘宝价格", "多少钱", "查一下价格", "销量", "店铺" or any product-related price inquiries. Can also perform other Taobao operations like viewing cart, checking orders, chatting with customer service, etc. - but NEVER payment or financial operations.
---

# Taobao Query Skill

Static analysis

No suspicious patterns detected.