Back to skill

Security audit

gtrk 视觉拆分派单

Security checks for vulnerabilities and agentic risk

Overview

The skill plans video visuals and limits further editing to the user's requested scope.

Keep a project backup because the workflow writes project annotations and can perform requested edits. Before requesting full production, review cloud costs and data handling. Cloud credentials are optional for cloud steps; the external CLI itself was not audited here.

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
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (2)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied commands modify existing clips through gtrk patch move and trim. They do not perform the declared split workflow, assign visual layers, or produce dispatch queues. These editing operations are materially different from the stated planning and dispatch purpose, which explicitly defers subsequent timeline assembly.

Content

No source excerpt is available for this finding.

The skill mandates English search descriptions and Chinese exclusions without user opt-in or a documented justification.

Low
Category
Output Handling
Confidence
98% confidence
Finding

The FILM_BROLL instructions explicitly require English scene descriptions for queries and Chinese text for exclude fields. This language requirement is reinforced at lines 173 and 304, but the skill neither offers a language choice nor explains a technical need for these restrictions.

Content

No source excerpt is available for this finding.