higgsfield-forensic-atlas
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill serves as an informational resource and configuration guide for the Higgsfield AI ecosystem. It does not contain executable scripts that perform automated tasks outside of documented API examples.
- [EXTERNAL_DOWNLOADS]: The skill references the official
@higgsfield/client/v2Node.js package and provides instructions for connecting to the official Higgsfield MCP server athttps://mcp.higgsfield.ai/mcp. These are recognized vendor-specific resources for the platform being documented. - [CREDENTIALS_UNSAFE]: Code examples demonstrate best practices by using environment variables (
process.env.HF_CREDENTIALS) for authentication instead of hardcoding secrets. It also notes that the MCP connector uses OAuth, further reducing the risk of credential exposure. - [INDIRECT_PROMPT_INJECTION]: While the skill documents tools that ingest untrusted external data (video and image URLs), it explicitly outlines ethical guidelines and biometric safety filters implemented by the platform to prevent abuse. There is no evidence of instructions that would cause the agent to bypass these guardrails.
Audit Metadata