tilebox-workflow-authoring
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill's instructions explicitly discourage insecure practices like hardcoding credentials or using local machine paths for dependencies. It promotes the use of standard authentication mechanisms and synchronized environments via tools like uv.
- [EXTERNAL_DOWNLOADS]: The documentation references official software repositories and package registries from trusted organizations, including the PyTorch foundation and Development Seed. These downloads are integral to the workflow's function and originate from well-known technology services.
- [INDIRECT_PROMPT_INJECTION]: The skill defines a processing surface where untrusted geospatial data and metadata are ingested from remote object storage and datasets (ingestion points). It manages this risk by establishing strict boundaries between task-serialized inputs, job-scoped caches, and durable storage (boundary markers). While the skill utilizes network and file-system operations through the Tilebox SDK and libraries like obstore (capability inventory), the instructions mandate that the agent validate external inputs and scientific preconditions during execution (sanitization).
Audit Metadata