Back to skill

Security audit

yu

Security checks for vulnerabilities and agentic risk

Overview

This is a Chinese academic paper review helper with a disclosed but imperfectly scoped text-sample feature, and it does not request system access, persistence, credentials, or external execution.

Install only if you want a Chinese-language blind-review style critique. Treat the modified text examples as illustrative editing suggestions, not as a substitute for your own authorship or institution-specific academic integrity rules.

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
  • 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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

This is a stronger form of the same issue: the skill's mandatory output requires the model to rewrite at least five passages from the manuscript despite a strict statement that rewriting is forbidden. Because the skill is designed for academic review, this inconsistency can materially enable inappropriate authorship assistance, policy evasion, and facilitation of academic integrity violations under the guise of 'examples.'

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill name, description, and all operating instructions are written to simulate a Chinese blind-review expert and implicitly constrain output behavior to Chinese academic review norms. There is no indication that users may choose another language or locale, nor any documented justification that the skill is region-specific only.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill contains an internal contradiction: it claims rewriting the manuscript is prohibited, yet the required output mandates at least five '修改后样例' derived from the user's original text. In practice this pushes the model to perform substantive text transformation on user content, which can bypass the stated safety boundary and create unauthorized editing/ghostwriting behavior rather than limited critique.

Content

No source excerpt is available for this finding.