The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a wide array of system diagnostic and management commands to perform host inspection and administration tasks.
Evidence: Employs pacman, systemctl, journalctl, bootctl, lsblk, lspci, findmnt, and other core Linux utilities throughout its workflow and reference files for gathering system state.
[PRIVILEGE_ESCALATION]: The skill provides instructions for performing high-privilege administrative operations using sudo, which is required for the intended functionality.
Evidence: Includes commands for system upgrades (sudo pacman -Syu), kernel/initramfs regeneration (sudo mkinitcpio -P, sudo dracut), bootloader installation (sudo bootctl install), and encryption management (sudo cryptsetup).
[EXTERNAL_DOWNLOADS]: The skill interacts with external package repositories and tracks software versions from official project sources.
Evidence: Facilitates the use of the Arch User Repository (AUR) via git clone and paru, and references official release pages on GitHub and GitLab for tracking versions of system components like systemd and PipeWire.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection due to its requirement to process untrusted or external system data.
Ingestion points: Reads system logs via journalctl, package metadata via pacman, and potentially malicious community-contributed PKGBUILD files during AUR workflows.
Boundary markers: Generally absent in command output gathering, although the skill includes a self-check requiring the agent to respect the "AUR trust boundary."
Capability inventory: High, including full package management, bootloader modification, and filesystem-level access.
Sanitization: Not explicitly defined for command outputs; relies on LLM guardrails for processing ingested text.

arch-linux