The Agent Skills Directory

[PROMPT_INJECTION]: The skill documents an automated installer pattern that is susceptible to indirect prompt injection through its input collection mechanism.
Ingestion points: In references/gum-prompts.md, the script templates use gum input and gum choose to collect values such as hostname, role, and ip directly from the user at runtime.
Boundary markers: The collected data is written to a flat YAML configuration file (site-config.yaml) without explicit delimiters or instructions for the agent to ignore embedded commands or control sequences.
Capability inventory: The generated configuration is subsequently consumed by high-privilege tools such as nixos-install (using the --flake argument) and systemd services defined in references/iso-building.md, which apply the configuration to the target system.
Sanitization: There is no evidence of input validation, escaping, or schema enforcement in the documented shell scripts before the user input is written to disk or interpolated into deployment commands.

nixos