gcp-cloud-run
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides patterns for processing external data via Google Cloud Pub/Sub and Cloud Storage event handlers. This introduces a surface for indirect prompt injection where maliciously crafted input data (e.g., message payloads or file metadata) could influence agent behavior or application logic if not properly sanitized.
- Ingestion points: Event processing logic in
references/detailed-guide.md, specifically theprocessPubSub,processStorageEvent, and/pubsubHTTP endpoint snippets. - Boundary markers: No explicit delimiters or instruction-ignoring warnings are present in the provided code snippets.
- Capability inventory: The skill defines actions for resource deployment (
gcloud), database interaction (pg,sqlalchemy), secret access (@google-cloud/secret-manager), and task queuing (google-cloud-tasks). - Sanitization: Input data is decoded and parsed as JSON, but there is no explicit validation logic to prevent command or prompt injection within the processed payloads.
- [EXTERNAL_DOWNLOADS]: The skill references and downloads official container images from Google's Container Registry (GCR), such as
gcr.io/cloud-builders/gcloud,gcr.io/cloud-builders/docker, andgcr.io/google.com/cloudsdktool/cloud-sdk. These are official tools from a well-known service provider used for building and deploying resources. - [COMMAND_EXECUTION]: Contains numerous implementation examples using the
gcloudCLI for deploying Cloud Run services and functions, creating secrets, and configuring infrastructure. These are standard operations for managing Google Cloud Platform environments.
Audit Metadata