higgsfield-youtube-thumbnail
Fail
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The bootstrap sequence downloads and executes an installation script for the higgsfield CLI from the author's official GitHub repository using a shell pipe.
- [COMMAND_EXECUTION]: Employs the Bash tool to interact with the higgsfield CLI for authentication, model status verification, and image generation. It uses a secure pattern of writing prompt content to temporary files and piping them to standard input to prevent shell injection from user-provided text.
- [PROMPT_INJECTION]: The skill processes untrusted user data, including video titles and headlines, which are interpolated into prompts for image generation models.
- Ingestion points: User-provided video topics, titles, and headlines in SKILL.md.
- Boundary markers: Uses a rigid 11-step prompt contract and specific IDENTITY LOCK blocks to constrain model behavior.
- Capability inventory: Utilizes the Bash tool for CLI execution and file writing operations.
- Sanitization: User inputs are written to temporary files before being passed to the CLI, effectively isolating the text from the shell command structure.
Recommendations
- HIGH: Downloads and executes remote code from: https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh - DO NOT USE without thorough review
Audit Metadata