higgsfield-soul-id

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFEREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill instructions include a bootstrap command that downloads and executes a shell script (install.sh) from the official Higgsfield AI GitHub repository using curl piped to sh to install the necessary CLI tools.\n- [EXTERNAL_DOWNLOADS]: Fetches the Higgsfield CLI installation script from the official vendor repository during the setup process.\n- [COMMAND_EXECUTION]: Utilizes the Bash tool to execute higgsfield CLI commands for account authentication, character model training, and AI image/video generation.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided character names and image file paths which are then used as arguments for training the Soul character model.\n
  • Ingestion points: Character names and local image file paths provided by the user in SKILL.md.\n
  • Boundary markers: No explicit delimiters or ignore instructions are defined for the input parameters.\n
  • Capability inventory: Execution of CLI commands through Bash and file-upload operations.\n
  • Sanitization: Input validation or sanitization measures are not defined in the skill instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 03:05 AM
Security Audit — agent-trust-hub — higgsfield-soul-id