Back to skill

Security audit

car-advisor

Security checks for vulnerabilities and agentic risk

Overview

This skill is a China-market car research helper that mainly instructs the agent to search public vehicle sources, with no executable code or persistence.

Install this if you want a Chinese-market vehicle comparison assistant that uses live web sources. Be aware it may favor Chinese sources and current China-market models, and do not follow the documented HTTP APPKEY API example without switching to HTTPS and safer secret handling.

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
references/data-sources.md:100
Finding

API Key Exposed Through Plaintext HTTP Query Parameter

Content
View full analysis

Vulnerability Details

File Location: references/data-sources.md, lines 100-105
Vulnerability Type: API credential exposure through plaintext transport and URL query parameters
Risk Level: Medium

Vulnerable Code

markdown
### 聚合数据 juhe.cn(个人可用)

- 接口:`http://op.juhe.cn/onebox/car/query`
- 参数:`key={APPKEY}&car={车型名}`
- 费用:按次收费,个人友好
- 申请:https://www.juhe.cn

Technical Analysis

The integration documentation instructs users to call an API over unencrypted HTTP and place the API key in the URL query string. Plaintext HTTP does not provide transport confidentiality or server authentication, so network intermediaries can observe or modify the request.

Even if transport encryption were added, credentials in query strings may be retained in application and proxy access logs, monitoring platforms, browser history, debugging output, and other URL-processing infrastructure. An exposed APPKEY could then be replayed independently of the legitimate client.

This is documentation rather than executable code, so exploitation requires a user or implementation to follow the insecure integration guidance.

Attack Path

  1. A developer follows the documented API example.
  2. The application sends a request to http://op.juhe.cn/onebox/car/query?key={APPKEY}&car={vehicle}.
  3. A network intermediary, compromised gateway, or logging component captures the complete URL.
  4. The attacker extracts the API key from the key parameter.
  5. The attacker submits requests using the stolen key until it is revoked or its quota is exhausted.

Impact Assessment

An attacker could obtain the exposed third-party API credential and use the privileges assigned to that key. The likely scope includes unauthorized API requests, quota exhaustion, possible usage charges, and access to any API operations authorized for the key. Vehicle query contents may also be disclosed.

This issue does not itself provide local code execution, host access, privilege escalation, ...[truncated 42 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace the plaintext endpoint with the provider's verified HTTPS endpoint.
  2. Use an Authorization header or another provider-supported credential mechanism instead of placing the API key in the query string.
  3. If the provider supports only query-based authentication, require HTTPS and configure applications, proxies, analytics systems, and monitoring tools to redact the key parameter.
  4. Store the API key in a secret manager or protected environment variable rather than source files or command-line arguments.
  5. Apply least-privilege restrictions, quotas, expiration, and source-IP or domain restrictions where supported.
  6. Rotate any key that may already have been transmitted through the documented HTTP endpoint.
  7. Update the example to show secure secret handling and explicitly warn against logging credentials.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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 (2)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill declares that it must trigger for essentially any question involving a specific car model or attribute, including cases where the user did not ask for comparison or live lookup. This overly broad routing can cause unnecessary tool invocation, increase exposure of user queries to external retrieval, and crowd out safer or more appropriate general-answer paths.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill content and prescribed behavior are entirely Chinese-centric without any opt-in, locale detection, or fallback behavior, which can force responses or retrieval patterns mismatched to the user's language. This can degrade safety and reliability by causing misunderstanding, incorrect source selection, or unintentional disclosure to region-specific sites when the user expected another locale.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.