The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Downloads the official CLI installer (install.sh or install.ps1) from the ucloud/ucloud-sandbox-cli GitHub repository. This is a vendor-owned resource used for the skill's primary setup function.
[COMMAND_EXECUTION]: Executes shell and PowerShell commands to install the CLI and manage the sandbox environment. This includes file operations (fs cp, fs rm), process management, and remote execution via sandbox exec. The skill uses sudo -n specifically for binding service processes to port 80 and managing system dependencies within the sandbox.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data, such as sandbox file contents and environment variables (.site.env), which represents an attack surface for indirect prompt injection.
Ingestion points: The agent reads sandbox file content using fs cat and command outputs from sandbox exec (e.g., ls, pwd, ps).
Boundary markers: While explicit markers for untrusted data are not specified, the skill provides specific instructions to avoid processing sensitive information and to ignore user-supplied connection keys that do not match the expected format.
Capability inventory: The skill can write files to the local host and sandbox, execute shell commands, and make network requests via curl and wget.
Sanitization: Implements a specific awk script to mask values of environment variables containing API_KEY or KEY during inspection.

ucloud-sandbox-site