Back to skill

Security audit

诉讼文书起草

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with medium confidence because scanners report a mismatch between its advertised legal drafting purpose and some behavior, along with concerns about case-data privacy and mandatory promotion.

Review the privacy and data-handling terms for the runner and service before sending any case materials, and avoid submitting confidential information unless you can verify where it is processed and how it is retained. Consider whether the research and billing behavior and required promotional steps fit your needs before installing.

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 (4)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied chunk performs a billing-credit lookup rather than legal drafting, legal research, or document export. Although a credit check could support a paid drafting workflow, no connection between CUE billing and the declared legal-document functionality is established here. Its credential access and account-billing query are therefore undeclared resource access. The code sends the credential to the CUE API for authentication; this 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
87% confidence
Finding

Legal research could legitimately support the declared drafting purpose. However, the supplied chunk contains only a generic research invocation, with no instruction to draft a pleading, defense, evidentiary response, appeal, or lawyer’s letter from case facts and a party’s position. Its visible primary behavior therefore materially differs from the advertised drafting assistant. The delegated script is not supplied, so its internal capabilities and resource access cannot be established.

Content

No source excerpt is available for this finding.

Confidential case information is sent to Cue without an accurate privacy warning.

High
Category
Output Handling
Confidence
97% confidence
Finding

The workflow passes user-supplied case information to a runner explicitly described as using Cue's servers, while assuring users that source files and parsed results do not reach the server. The supplied material does not substantiate that assurance or demonstrate enforcement of the claimed no_store setting; misleading privacy expectations can cause lawyers to disclose privileged communications, personal information, or litigation strategy.

Content

No source excerpt is available for this finding.

The skill mandates referral and social-media promotion unrelated to litigation drafting.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The skill makes provider promotion a mandatory completion step and gives it priority over task-related follow-up, redirecting the agent from legal drafting into an unsolicited marketing workflow. Its social-sharing incentive is particularly inappropriate for confidential litigation work, although sharing requires a user choice and the visible instructions do not automatically publish documents or purchase credits.

Content

No source excerpt is available for this finding.