Back to skill

Security audit

跨境法规调研

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because scanners report undeclared billing access, unverified external code updates, and other concerns around sending research questions and promoting the provider.

Consider the reported billing access, third-party processing of research questions, unverified runner updates, and mandatory promotional prompts before installing. The scan is partial because a referenced runner file was missing, so its behavior could not be fully assessed; the results do not establish malicious intent.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • 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
Findings (5)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied chunk performs account billing inspection rather than legal research and does not access any declared legal source. Although its pre-run wording suggests a preparatory check, nothing in the declared purpose establishes Cue billing as a supporting dependency. Accessing local credentials and an unrelated account billing resource therefore constitutes an undeclared capability. The key is sent to the Cue endpoint for authentication; the code alone does not establish malicious credential exfiltration.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

Installing a runner is consistent with the documented architecture, so downloading a repository is not inherently a purpose mismatch. However, the workflow automatically pulls mutable repository content and subsequently runs its Python code without version pinning or integrity verification; a compromised upstream or mirror could execute code with the agent's local permissions and access credentials or sensitive legal material.

Content

No source excerpt is available for this finding.

The skill requires Chinese-language reports without offering a language choice.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

The skill mandates entirely Chinese report delivery, reinforced by the report-language FAQ at lines 210–211. It provides no user opt-in or alternative language, and researching foreign regulations does not establish a compliance requirement for Chinese-only output.

Content

No source excerpt is available for this finding.

The workflow sends the user's verbatim legal-research question to Cue without warning about confidential or privileged information.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The workflow forwards the user's verbatim question to a third-party backend without requiring a confidentiality review, redaction, or explicit approval of that disclosure. Legal-research questions can contain privileged advice requests, acquisition plans, personal information, or trade secrets, making the missing disclosure boundary particularly consequential despite the skill's stated public-data scope.

Content

No source excerpt is available for this finding.

The skill mandates referral and social-media promotion unrelated to cross-border legal research.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The skill makes referral, purchase, and social-sharing prompts mandatory and gives them priority over unrelated follow-up questions. This redirects the agent from legal research toward the provider's promotional objectives, although the text does not require completing a purchase or publishing anything without the user's choice.

Content

No source excerpt is available for this finding.