higgsfield-soul-id
Pass
Audited by Gen Agent Trust Hub on Aug 6, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill instructions include a bootstrap step to install the '@higgsfield/cli' package globally using 'npm install -g'. This is the standard method for deploying the vendor's command-line interface tools.
- [INDIRECT_PROMPT_INJECTION]: The skill constructs shell commands by interpolating user-provided character names and file paths, creating a surface for potential command injection if the inputs are not correctly sanitized by the agent.
- Ingestion points: User-provided character names and local photo paths in SKILL.md.
- Boundary markers: The workflow examples use double quotes for the character name argument, but photo paths are interpolated directly without explicit delimiters or sanitization warnings.
- Capability inventory: The skill uses the 'Bash' tool to execute all CLI commands for creating, waiting for, and listing identity models.
- Sanitization: There are no explicit instructions for the agent to filter, validate, or escape the external content before interpolation into the bash commands.
Audit Metadata