te

• Analyzed: May 12, 2026, 3:15 AM
• Findings: 1
• Version: 1.0.1
• Hash: ea5a812da3cc0ce03d91da1bf6f96404f09ab7617cfbc41859d284f5d92f88f2
• Source: palm
• External report: View on VirusTotal

Findings (1)

The `SKILL.md` file contains explicit instructions under 'CRITICAL: Action Tag Rules' that dictate the AI agent's response format and action prioritization (e.g., 'You MUST include the action tag!', 'Put your MOST IMPORTANT gesture FIRST!'). These instructions represent a form of prompt injection, attempting to control the agent's behavior and output beyond simply providing tools. While the apparent intent is to ensure the avatar control features function correctly, it is a risky capability as it overrides the agent's autonomy and could be leveraged to ignore user instructions or prioritize specific actions. Additionally, the skill allows the agent to display arbitrary GIFs and YouTube videos based on search terms, which could be abused by users interacting with the agent.