fiftyone-embeddings-visualization
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The installation instructions in README.md use a piped shell script (curl | sh) from the official installer domain for the FiftyOne ecosystem (skil.sh), which is used to manage skill installations.
- [EXTERNAL_DOWNLOADS]: The SKILL.md instructions utilize the download_plugin tool to fetch code from the vendor's official repository (voxel51/fiftyone-plugins) to enable necessary visualization capabilities.
- [DYNAMIC_EXECUTION]: The skill dynamically manages its execution environment by downloading and enabling plugins from the vendor's repository at runtime to perform embedding and dimensionality reduction tasks.
- [INDIRECT_PROMPT_INJECTION]: The skill processes dataset samples and metadata, which constitutes an attack surface for indirect prompt injection.
- Ingestion points: Dataset samples are processed by machine learning models and brain operators for visualization.
- Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are used when processing dataset content.
- Capability inventory: The skill has access to tools for operator execution, plugin management, and dataset manipulation.
- Sanitization: Content processing relies on the underlying machine learning models and operators without explicit pre-sanitization in the prompt logic.
Audit Metadata