The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents extensive use of the lxc and incus CLI tools for system-level operations, including launching containers, executing commands inside containers, and modifying host-level network and storage configurations.
[EXTERNAL_DOWNLOADS]: Fetches an installation script from the well-known Docker domain (get.docker.com) for container setup.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: Demonstrates an automated installation pattern that pipes a remote script from a well-known source directly to a shell (curl | sh) for Docker installation.
[PRIVILEGE_ESCALATION]: Documents the use of privileged containers (security.privileged=true) and group membership modifications (usermod -aG lxd), providing necessary security warnings about the implications of sharing the host kernel namespace.
[INDIRECT_PROMPT_INJECTION]: Identifies an attack surface for indirect prompt injection:
Ingestion points: User-provided container names and cloud-init YAML configurations in references/common-patterns.md.
Boundary markers: None present in the command templates to differentiate between instructions and untrusted data.
Capability inventory: High-privilege operations including lxc exec, lxc file push, and lxc config set are available across the provided scripts.
Sanitization: No explicit sanitization or validation of input data is described in the reference patterns.

lxc-lxd