The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes remote domain and subnet lists from external community sources, creating a potential attack surface for indirect instructions.
Ingestion points: Remote lists specified in remote_domain_lists and remote_subnet_lists (referenced in references/operations.md).
Boundary markers: The skill includes a 'Safety Note' in SKILL.md explicitly stating that remote domain/subnet lists and community data should not be considered automatically trusted.
Capability inventory: The orchestration tool manages core system networking including sing-box, dnsmasq, and nftables, and performs shell-based configuration (referenced in references/operations.md and references/repo-map.md).
Sanitization: The documentation notes that proxy strings and inputs are validated by the frontend and shell-based backend helpers (references/proxy-strings.md).- [EXTERNAL_DOWNLOADS]: The guide provides instructions to download and execute an installation script and a configuration file from an external repository.
Evidence: references/operations.md describes fetching install.sh and podkop.config from raw.githubusercontent.com/itdoginfo/podkop/ using curl.- [COMMAND_EXECUTION]: The skill requires the execution of multiple system commands to manage the router's configuration and service lifecycle.
Evidence: references/operations.md details the use of commands like podkop start, podkop reload, and podkop global_check to control system-level routing services.

podkop-openwrt-guide