fiftyone-code-style
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION] (HIGH): The skill directs the agent to search and read content from an external repository (https://github.com/voxel51/fiftyone) to avoid redundant code. This creates a significant indirect prompt injection surface.
- Ingestion points: Remote search results from
https://github.com/voxel51/fiftyonereferenced inSKILL.md. - Boundary markers: Absent. No instructions are provided to the agent to delimit or ignore instructions embedded in the external code it reads.
- Capability inventory: The agent is tasked with writing and implementing Python code based on this external content, which constitutes a high-privilege write capability.
- Sanitization: Absent. There is no validation or filtering of the content retrieved from the external repository.
- [EXTERNAL_DOWNLOADS] (MEDIUM): The skill promotes the use of
fou.ensure_package(), which functions as a runtime package manager to download and install third-party dependencies. - Evidence:
SKILL.mdcontains code examples usingfou.lazy_import("open3d", callback=lambda: fou.ensure_package("open3d"))andfou.ensure_import("pycocotools"). - [REMOTE_CODE_EXECUTION] (MEDIUM): By using
ensure_package, the skill enables the download and subsequent execution (via import) of arbitrary code from external package registries at runtime. - Evidence: The use of
fou.lazy_importwith anensure_packagecallback inSKILL.mdspecifically triggers this behavior when the dependency is missing.
Recommendations
- AI detected serious security threats
Audit Metadata