The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill promotes an installation method that pipes a remote shell script directly into a root shell (curl|bash). This pattern allows a remote server to execute arbitrary commands on the user's host system without prior review.
Evidence: curl -fsSL https://raw.githubusercontent.com/MengMengCode/CLICD/main/install.sh | sudo sh
[PRIVILEGE_ESCALATION]: The skill frequently instructs the agent to use sudo for installation, service management, and network configuration, providing a wide surface for administrative abuse if the downloaded code is compromised.
Evidence: Multiple commands prefixed with sudo, including the piping of a remote script to a privileged shell.
[EXTERNAL_DOWNLOADS]: The skill fetches resources from a third-party GitHub repository (MengMengCode) that does not match the stated author and is not part of a recognized trusted organization.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted system data and API responses without sufficient security boundaries or sanitization, creating a potential vulnerability to data-driven instruction injection.
Ingestion points: clicd container list, clicd user list, and security alert logs.
Boundary markers: Absent. There are no instructions for the agent to distinguish between trusted commands and untrusted data outputs.
Capability inventory: Extensive capabilities including arbitrary CLI command execution, file system management via snapshots, and networking modifications via iptables.
Sanitization: Absent. The skill does not provide mechanisms for validating or escaping external data before it is processed by the agent.

clicd-virtualization-panel