media-higgsfield-core
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a framework for processing external data such as user prompts and media identifiers.
- Ingestion points: Ingests user data via prompt and medias fields in generation tool parameters as defined in references/call-schema.md.
- Boundary markers: Utilizes a structured Socratic slots system in references/interview-schema.md to maintain context boundaries during data collection.
- Capability inventory: Capabilities are restricted to authorized image and video generation tools.
- Sanitization: The skill mandates a hard user approval gate for all credit-consuming tasks and includes logic to process and report nsfw content filter failures, mitigating risks of automated or unsafe generation.
- [DYNAMIC_EXECUTION]: The skill uses a runtime namespace resolution pattern to identify the correct tool prefix for environment compatibility.
- Evidence: references/call-schema.md describes selecting between the mcp__higgsfield__ and mcp__claude_ai_higgsfield__ prefixes using ToolSearch, which is a standard configuration-level tool discovery mechanism rather than arbitrary code execution.
Audit Metadata