api-manager-base

Pass

Audited by Gen Agent Trust Hub on Sep 22, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill establishes a declarative pattern for managers that ingest data from external sources, such as file uploads and user-provided metadata in the MediaAssetManager. This design creates a standard attack surface for indirect prompt injection where untrusted content could potentially influence agent behavior if not properly handled by downstream consumers.
  • Ingestion points: MediaAssetManager.create and BaseManager.update in SKILL.md process user-supplied inputs and file objects.
  • Boundary markers: The pattern relies on external schema validation but does not explicitly define prompt-level delimiters within the base class logic.
  • Capability inventory: The managers possess capabilities to perform database writes via Prisma and object storage operations via AWS S3.
  • Sanitization: The implementation utilizes pumice.js for input schema validation, though content-level sanitization of file bodies is deferred to specific model implementations.
  • [EXTERNAL_DOWNLOADS]: The implementation examples reference standard Node.js dependencies including @aws-sdk/client-s3 for storage operations and image-size for metadata extraction. It also mentions pumice.js for data validation, which serves a legitimate functional role in the described architecture.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 22, 2026, 08:22 PM