heygen-avatar

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill provides a fallback instruction to install the HeyGen CLI by downloading a script from the vendor's official domain (https://static.heygen.ai/cli/install.sh). This is a standard setup procedure for the vendor's own tooling.
  • [REMOTE_CODE_EXECUTION]: The skill includes a command that pipes a remote installer script directly to bash (curl -fsSL https://static.heygen.ai/cli/install.sh | bash). This is presented as a visible instruction for the user to configure their environment when automated MCP tools are unavailable.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests identity traits and appearance descriptions to generate prompts for the HeyGen avatar API, creating an injection surface.
  • Ingestion points: Character details are read from workspace files like SOUL.md and IDENTITY.md, as well as from natural language descriptions provided by the user.
  • Boundary markers: The skill uses a multi-phase logic to separate identity extraction from API execution, though no explicit delimiters are defined for the prompt construction.
  • Capability inventory: The skill has access to file system operations (Write for avatar configuration files), shell commands via Bash (used for creating symlinks), and the mcp__heygen__* tool suite for API calls.
  • Sanitization: The skill maps free-form identity attributes (age, gender, ethnicity, style) to a strictly defined set of HeyGen API enums to ensure structured data is sent to the backend.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 10:02 AM
Security Audit — agent-trust-hub — heygen-avatar