The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill describes a workflow to download an installation script from a remote repository and execute it with shell privileges.
Evidence: references/operations.md contains a command sequence to fetch install.sh from GitHub and run it via sh /tmp/podkop-install.sh.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface where remote domain and subnet lists are processed to generate system routing configurations, which could lead to routing bypass or manipulation if the external lists are poisoned.
Ingestion points: remote_domain_lists, remote_subnet_lists, and community_lists defined in references/operations.md.
Boundary markers: The skill includes a "Safety Note" in SKILL.md advising that remote lists are external inputs and should be verified.
Capability inventory: The orchestration scripts have the capability to modify dnsmasq, sing-box, and nftables rules.
Sanitization: Validation logic is mentioned in helpers.sh and frontend validators, but runtime data remains external.
[COMMAND_EXECUTION]: The skill uses shell-based orchestration to manage system-level services and configurations on OpenWrt.
Evidence: podkop/files/usr/bin/podkop is a large shell script (~2700 lines) that manipulates UCI, dnsmasq, and firewall settings.
[EXTERNAL_DOWNLOADS]: The skill fetches installation scripts and configuration templates from the project's official GitHub repository.
Evidence: references/operations.md points to raw.githubusercontent.com/itdoginfo/podkop/ for resources.

podkop-openwrt-guide