gke-ai-troubleshooting-jobset-interruption

Pass

Audited by Gen Agent Trust Hub on Jul 21, 2026

Risk Level: SAFECOMMAND_EXECUTIONCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Diagnostic procedures involve executing shell commands with curl to interact with Google Cloud APIs, requiring the BypassSandbox: true capability.
  • [CREDENTIALS_UNSAFE]: The workflow uses gcloud auth print-access-token to retrieve and transmit the user's active session token to official Google Cloud Monitoring endpoints.
  • [EXTERNAL_DOWNLOADS]: Fetches metrics and diagnostic data from the official Google Cloud Monitoring service.
  • [PROMPT_INJECTION]: The skill ingests user inputs (project ID, cluster name, workload name) which are interpolated into shell commands and log filters.
  • Ingestion points: Project and cluster metadata provided by the user in Step 0.
  • Boundary markers: Absent; inputs are used directly in string templates.
  • Capability inventory: Shell command execution and log retrieval.
  • Sanitization: Absent; the instructions do not specify validation or escaping for user-provided strings.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 21, 2026, 05:58 PM
Security Audit — agent-trust-hub — gke-ai-troubleshooting-jobset-interruption