The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download the Huawei Cloud KooCLI (hcloud) installer from an official vendor-controlled bucket at hwcloudcli.obs.cn-north-4.myhuaweicloud.com. This is a legitimate download from the service provider's infrastructure.
[REMOTE_CODE_EXECUTION]: The installation guide for the CLI backend includes a command pattern that fetches and executes a remote shell script (hcloud_install.sh) using curl | bash. While this is the official installation method provided by the vendor, it represents the execution of remote code.
[COMMAND_EXECUTION]: The skill uses the hcloud CLI and a bundled Python script (swr_instance_helper.py) to perform cloud management tasks. These include sensitive operations such as creating access credentials, managing IP whitelists for public endpoints, and deleting entire registry instances.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from external API responses and interpolates user-controlled strings into shell commands and SDK calls.
Ingestion points: The agent ingests data from instance lists, repository details, and job logs (hcloud SWR ListInstanceJobs) which could contain attacker-controlled metadata. User-supplied parameters like --name and --description are also used.
Boundary markers: The instructions do not define delimiters or specific warnings to prevent the agent from following instructions embedded in the metadata of the cloud resources it manages.
Capability inventory: The skill has high-tier capabilities, including network communication via the SDK, the ability to generate and retrieve long-term credentials (CreateInstanceLtCredential), and the power to delete storage and DNS records.
Sanitization: The skill relies on standard shell quoting and CLI argument parsing; it does not explicitly sanitize or validate the content of remote API responses before they are processed by the agent.

huawei-cloud-swr-enterprise-instance