Baoyu Image Cards

• Outcome: Review
• Latest audit: May 28, 2026 at 2:25 PM UTC
• Version:
  1.117.2

Overview

This is mostly a normal image-card generator, but it includes reusable prompt text that pushes image tools not to refuse sensitive or copyrighted figure requests.

Review or change the prompt-safety language before installing. The local file-writing and preference behavior is disclosed and proportionate, but this skill may steer image backends toward stylized substitutes for sensitive or copyrighted figures instead of allowing normal refusal or policy handling.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (4)

Context-Inappropriate Capability

Medium

• Confidence: 97% confidence
• Finding: The prompt explicitly tells the agent to avoid refusing requests involving sensitive or copyrighted figures by generating 'stylistically similar alternatives.' That weakens safety boundaries and can be used to circumvent model refusals or policy protections around sensitive persons and copyrighted characters, which is not necessary for a generic infographic-card workflow.

Context-Inappropriate Capability

Medium

• Confidence: 95% confidence
• Finding: The screen-print override repeats the same unsafe pattern by instructing the system to create stylistically similar silhouettes for sensitive or copyrighted figures. Even framed as an art-direction rule, it encourages transformation-based evasion of refusal safeguards and IP/sensitive-person restrictions.

Natural-Language Policy Violations

Medium

• Confidence: 96% confidence
• Finding: This instruction directly discourages refusal for a risky content category. In a prompt-assembly component, that means unsafe behavior can be propagated into every generated prompt, increasing the chance that downstream image models are nudged to produce content they would otherwise reject.

Ssd 1

Medium

• Confidence: 94% confidence
• Finding: The file disguises refusal suppression as a stylistic instruction ('create stylistically similar alternatives'), which makes the unsafe behavior seem harmless and easier to miss in review. That concealment increases risk because it normalizes policy evasion in a reusable prompt template for broad user inputs.

VirusTotal

VirusTotal engine telemetry is currently stale for this artifact.