gcp-cloud-functions
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The code templates create a surface for indirect prompt injection by processing external data from GCP triggers.
- Ingestion points: The skill ingests untrusted data from HTTP request arguments (request.args) and event payloads from Pub/Sub and Cloud Storage (cloud_event.data).
- Boundary markers: The examples do not use delimiters or explicit 'ignore instructions' markers for untrusted data.
- Capability inventory: The functions demonstrate capabilities to write to Firestore databases and perform image analysis via the Vision API.
- Sanitization: The templates show direct interpolation of external inputs into response strings and database keys without sanitization or validation.
- [COMMAND_EXECUTION]: The skill relies on gcloud CLI commands to manage GCP infrastructure. These include service enablement, resource creation, and function deployment. The instructions provide warnings about cost, scope, and potential service disruption.
- [EXTERNAL_DOWNLOADS]: The skill references official Google Cloud libraries and the gcloud SDK. These dependencies are from a well-known service provider and are necessary for interacting with Google Cloud services.
Audit Metadata