agent-platform-alert-configuration
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The scripts/gather_agent_info.py script executes local system commands including gcloud and bq to discover project resources such as Reasoning Engines and BigQuery datasets. The implementation uses list-based command construction for safe parameter passing.
- [EXTERNAL_DOWNLOADS]: The skill interacts with official Google Cloud APIs (logging.googleapis.com, monitoring.googleapis.com, observability.googleapis.com) to retrieve metadata and telemetry. These operations target well-known, trusted services and use standard OAuth2 authentication via the google-auth library.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided identifiers like agent names and interpolates them into PromQL and SQL query templates. While these queries are limited to telemetry filtering within the Google Cloud Monitoring and Logging APIs, the absence of strict input sanitization creates a potential surface for query injection if malicious resource names are encountered.
- Ingestion points: User-provided agent-name and project-id in SKILL.md instructions.
- Boundary markers: SKILL.md includes explicit instructions to only use projects provided by the user.
- Capability inventory: Resource discovery via gcloud/bq, telemetry retrieval via requests to GCP APIs, and file-writing for Terraform HCL.
- Sanitization: None observed for resource identifiers interpolated into query templates.
Audit Metadata