The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's documentation (references/kubectl-cce.md) instructs the user to download a binary plugin from an unverified external source: https://github.com/pancake0001/kubectl-cce-plugin/releases/download/v0.1.0/kubectl-cce_0.1.0_linux_amd64.tar.gz. As this repository belongs to an unverified third-party account, the security and integrity of the binary cannot be guaranteed.
[COMMAND_EXECUTION]: The skill invokes external command-line tools (hcloud and kubectl) using subprocess.run() in scripts/huawei_cloud/common.py and scripts/huawei_cloud/kubectl_client.py. These commands are passed sensitive credentials (AK/SK) and user-supplied parameters, which increases the risk of command injection or accidental exposure if logs are mismanaged.
[CREDENTIALS_UNSAFE]: The skill retrieves sensitive cryptographic material by reading TLS certificates from Kubernetes Secrets in scripts/huawei_cloud/cce.py. Additionally, while scripts/huawei_cloud/common.py includes redaction functions, the skill still handles raw AK/SK credentials in memory and environment variables.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8).
Ingestion points: Data from external metric queries (AOM Prometheus) and Kubernetes resource metadata (Pods, Services, Ingresses) is retrieved in scripts/huawei_cloud/cce_metrics.py and scripts/huawei_cloud/kubectl_client.py and ingested into the agent context.
Boundary markers: The skill lacks explicit delimiters to distinguish untrusted external data from the agent's internal instructions.
Capability inventory: The skill possesses the ability to execute CLI commands and perform authenticated network requests to cloud services.
Sanitization: No sanitization or validation is applied to data fetched from the cluster environment before it is processed for status reporting.

huawei-cloud-cce-metric-analyzer