virtual-input

Pass

Audited by Gen Agent Trust Hub on Jun 12, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill features an indirect prompt injection surface by ingesting untrusted text from the game environment during UI discovery.
  • Ingestion points: The Luau scripts provided in SKILL.md and commands/vi-walkthrough.md iterate through the PlayerGui tree and extract the Text property of labels, buttons, and text boxes to identify elements.
  • Capability inventory: The skill utilizes execute_luau for script execution, screen_capture for visual feedback, and provides instructions for writing files to the local disk.
  • Sanitization: There is no evidence of filtering or sanitization of the extracted UI text before it is returned to the agent context.
  • Boundary markers: The instructions do not specify the use of delimiters or warnings to ignore instructions that might be embedded within the UI text.
  • [COMMAND_EXECUTION]: The skill relies on dynamic script generation and execution to perform its primary functions.
  • Evidence: All command files, including vi-click.md, vi-key.md, and vi-type.md, instruct the agent to generate Luau code blocks and execute them via the execute_luau tool.
  • Context: This dynamic execution is targeted at the Roblox engine's UserInputService to inject virtual inputs, which is the core functionality of the skill.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 12, 2026, 08:32 PM
Security Audit — agent-trust-hub — virtual-input