huawei-cloud-ucs-cluster-onboarding-manager

Pass

Audited by Gen Agent Trust Hub on Jun 17, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it ingests untrusted Kubernetes configuration files provided by users or external clusters.
  • Ingestion points: Untrusted data enters the agent context via the --metadata.annotations.kubeconfig parameter during cluster registration (documented in SKILL.md and references/task-cluster-registration.md).
  • Boundary markers: No explicit delimiters or boundary markers are defined to isolate this untrusted content from the agent's instructions.
  • Capability inventory: The skill has powerful management capabilities, including the ability to delete clusters, modify fleet groups, and retrieve credentials (e.g., hcloud UCS DeleteCluster, hcloud UCS JoinGroup, hcloud UCS CreateClusterKubeconfig).
  • Sanitization: There is no evidence of sanitization or strict schema validation for the input YAML before it is interpolated into shell commands.
  • [COMMAND_EXECUTION]: The skill performs cloud management operations by executing shell commands via the hcloud CLI tool.
  • Evidence: All primary operations, such as hcloud UCS RegisterCluster, hcloud UCS ShowCluster, and hcloud UCS DeleteCluster, rely on executing shell commands with user-provided parameters.
  • Risk: This reliance on shell interpolation requires the agent to carefully validate all user-supplied inputs to prevent command injection vulnerabilities.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 17, 2026, 09:10 AM
Security Audit — agent-trust-hub — huawei-cloud-ucs-cluster-onboarding-manager