Back to skill

Security audit

Subclip Local Asset Creation

Security checks for vulnerabilities and agentic risk

Overview

The skill received a suspicious verdict because scanners found concerns about authorizing cloud image generation despite local-only framing, while no malicious behavior was reported; scan coverage was partial.

Before installing, consider whether you are comfortable with prompts or images being sent to an external provider, stored in the cloud, and potentially using your AI credits. The scan covered the listed components but was marked partial because a referenced local path was missing, so the evidence is incomplete.

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)

The local-only creation description conflicts with instructions authorizing cloud generation.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The manifest describes creating and revising images locally, but L10 authorizes external model providers and Subclip cloud generation, explicitly involving user AI credits and cloud storage. Cloud generation may be selected simply because its models are considered the best fit, rather than because the user requested an exception to local processing.

Content

No source excerpt is available for this finding.

The workflow permits external generation providers without warning that user content may leave the local environment.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The skill authorizes any accessible generation provider and permits cloud selection based on model suitability without requiring approval before sending user material externally. In image-editing workflows, this could disclose private reference images or confidential prompts to a third party, particularly when the local-creation framing leads users to expect on-device processing. No malicious destination or actual transmission is demonstrated, so this is an unsafe authorization default rather than evidence of deliberate exfiltration.

Content

No source excerpt is available for this finding.