extension-object-storage

Pass

Audited by Gen Agent Trust Hub on Oct 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill provides a surface for ingesting untrusted data through file uploads.
  • Ingestion points: User-provided files (images, videos, documents) are processed via the handleUpload function and stored in the backend using the Storage.ExternalBlob type.
  • Boundary markers: No explicit prompt boundary markers or specific warnings for the agent to ignore embedded instructions within file content are present.
  • Capability inventory: The skill enables file storage, direct URL generation for inline display, and byte-level access for downloads. The provided examples do not show content being executed as code.
  • Sanitization: The documentation recommends identifying file types using the backend-stored filename or mimeType fields rather than the generated URL, but does not specify sanitization for the binary content itself.
  • [EXTERNAL_DOWNLOADS]: The skill requires the installation of specific packages from the vendor ecosystem.
  • References the caffeineai-object-storage package (version ~1.1.0 or 0.1.2) for the mops package manager.
  • References the @caffeineai/object-storage package for the npm package manager.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 7, 2026, 04:08 PM