The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs users to download mod files from an unverified personal GitHub Pages site (https://maglin-jenebellah.github.io). This source is not a trusted organization or well-known service, which is a common vector for distributing malicious software.
[REMOTE_CODE_EXECUTION]: Users are instructed to manually install binary files (DeepSynergy.dll, SyncEngine.dll, NetworkLayer.dll) into a game's plugin directory. These DLLs are loaded and executed by a hooking framework at runtime, allowing for arbitrary code execution on the user's system.
[COMMAND_EXECUTION]: The skill provides instructions for the user to execute shell commands for session management and networking.
[PRIVILEGE_ESCALATION]: The troubleshooting section explicitly instructs users to use sudo for firewall modifications (sudo ufw allow 7777/udp) and process management (sudo lsof -i :7777). Executing instructions from an untrusted source with root privileges is a high-risk activity.
[INDIRECT_PROMPT_INJECTION]: The skill includes an AI-integrated narrative feature that processes game events. 1. Ingestion points: The /api_narrate command takes event descriptions which could potentially come from untrusted peer data in a multiplayer session. 2. Boundary markers: There are no markers or warnings provided to the agent to treat this data as untrusted content. 3. Capability inventory: The mod is implemented as binary DLLs with full access to the Unity runtime and system resources. 4. Sanitization: No sanitization or filtering of the event data is described before it is sent to the LLM.

subnautica-ii-coop-deepsynergy-mod