higgsfield
Fail
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSOBFUSCATION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The documentation includes installation instructions that use piped shell execution to download and run scripts directly.
- Evidence:
curl -fsSL https://raw.githubusercontent.com/Koushik890/higgsfield-api-skill/main/install.sh | shin README.md. - Evidence:
irm https://raw.githubusercontent.com/Koushik890/higgsfield-api-skill/main/install.ps1 | iexin README.md. - [DYNAMIC_EXECUTION]: The skill uses dynamic module loading by encoding code into a Base64 data URI at runtime.
- Evidence:
return import("data:text/javascript;base64," + Buffer.from(module).toString("base64"))inbin/agents.js. - [INDIRECT_PROMPT_INJECTION]: The skill ingests external data from the Higgsfield API and documentation sites.
- Ingestion points: Fetches model metadata and schemas from open.higgsfield.ai and dash.higgsfield.ai.
- Boundary markers: None explicitly defined for external API data ingestion.
- Capability inventory: Subprocess calls to Python scripts, file system writes for downloads and model syncs, and network communication with Higgsfield endpoints.
- Sanitization: Implements strict JSON schema validation for all model input parameters in
scripts/hf.py. - [OBFUSCATION]: The agent detection script uses Base64 encoding to dynamically load a generated JavaScript module.
- [EXTERNAL_DOWNLOADS]: The skill fetches model configuration and updates from remote Higgsfield servers and provides installation scripts hosted on GitHub.
Recommendations
- HIGH: Downloads and executes remote code from: https://raw.githubusercontent.com/Koushik890/higgsfield-api-skill/main/install.sh - DO NOT USE without thorough review
Audit Metadata