higgsfield-canvas
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFENO_CODE
Full Analysis
- [NO_CODE]: The skill consists entirely of markdown documentation and YAML metadata. It contains no scripts, binaries, or executable commands, posing no direct execution risk.
- [INDIRECT_PROMPT_INJECTION]: The skill describes a system (Higgsfield Canvas) designed to process user-supplied prompts and data across various generative models and LLM nodes. While this architecture naturally possesses an attack surface for indirect prompt injection, there is no evidence of malicious exploitation within the skill instructions itself.
- Ingestion points: Prompt nodes, LLM assistant nodes, and user-uploaded assets (references) as described in SKILL.md.
- Boundary markers: None mentioned in the descriptive text.
- Capability inventory: Image, video, and audio generation; LLM-based prompt shaping; and real-time collaboration (Shared Canvas).
- Sanitization: No specific sanitization or filtering logic is defined within the instructional content.
Audit Metadata