higgsfield-soul-id
Fail
Audited by Gen Agent Trust Hub on Jun 25, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: Provides a command to install the necessary CLI tool by piping a remote shell script directly to the shell (
curl -fsSL https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh | sh). The script is hosted by the vendor's official GitHub organization.\n- [COMMAND_EXECUTION]: Relies on the execution of multiple shell commands via thehiggsfieldCLI to manage user authentication, model training, and status polling.\n- [DATA_EXFILTRATION]: Instructs the agent to upload local image files provided by the user to the vendor's cloud service to facilitate identity-faithful model training.\n- [PROMPT_INJECTION]: Contains a surface for indirect prompt injection through the processing of untrusted user data.\n - Ingestion points: User-provided character names and local file paths for images in
SKILL.md.\n - Boundary markers: Instructions specify the name should be "One word", but no explicit shell escape sequences or delimiters are defined for file paths.\n
- Capability inventory: Extensive use of shell commands (
higgsfield soul-id create,higgsfield soul-id wait) that process the ingested data.\n - Sanitization: The agent is instructed to ensure the name is a single word, which provides a basic level of mitigation against complex shell injection payloads.
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