The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references downloads for operating system cloud images and installation ISOs from official sources including Debian, Ubuntu, Alpine Linux, the Fedora Project, and Rocky Linux. These downloads are required for virtualization tasks and are documented neutrally as they target well-known organizations.
[REMOTE_CODE_EXECUTION]: Infrastructure automation is performed using industry-standard tools like Terraform and Packer. Provider and plugin sources are established and necessary for the skill's primary management functions.
[CREDENTIALS_UNSAFE]: The skill explicitly enforces best practices for secret management, instructing users and agents to avoid hardcoding sensitive information. It recommends using API tokens with least privilege, SSH public keys, and environment variables (e.g., TF_VAR_proxmox_api_token) or protected files for password storage.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Processes external resources such as OS images, hypervisor configurations, and user-supplied templates (SKILL.md, references/proxmox.md).
Boundary markers: Includes explicit guidelines to use cloud-init and variable injection for configuration rather than manual post-boot changes.
Capability inventory: Comprehensive shell execution capabilities via qm, virsh, pvesh, and pct CLI tools for VM and container lifecycle management.
Sanitization: Recommends the use of verified SHA-256 digests for external files and the application of least-privilege principles for API access.
[COMMAND_EXECUTION]: Usage of administrative shell commands (e.g., lvm, qm, virsh) is standard for hypervisor management and is correctly limited to the skill's virtualization domain.

virtualization