The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The red launcher script downloads dependencies from GitHub repositories (github:getcolors/agent-network and github:getcolors/once) if they are not found in the local cache. These repositories are hosted on a well-known service and belong to the skill's authoring organization.
[COMMAND_EXECUTION]: The launcher script executes the bun install command via Bun.spawnSync to install dependencies into a localized cache directory.
[DYNAMIC_EXECUTION]: The script dynamically loads the downloaded code using import() from a computed path in the local cache directory. The path is derived from a hash of the hardcoded dependency specifiers.
[PERSISTENCE]: The skill modifies the host's ~/.ssh/config file to add and manage SSH aliases and IdentityFile entries for the nodes it provisions. This is an intended feature for managing remote infrastructure but constitutes a persistent modification to the host environment.
[INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection as it ingests user-provided configuration (colors.yml) and manages high-privilege cloud infrastructure.
Ingestion points: The skill reads configuration from colors.yml and environment variables in .envrc.private.
Boundary markers: The skill documentation mentions validation logic and refusal of mismatched or legacy infrastructure states.
Capability inventory: Capability to provision and destroy virtual machines on Vultr and DigitalOcean, modify provider firewall rules, manage DNS records on Cloudflare, and deploy Docker containers.
Sanitization: The skill includes validation checks during the build and create phases, reporting usage failures through specific exit codes.

package-agent-network-red