The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines patterns for a 100-player authoritative server, creating a surface where the agent (if running the server logic) processes untrusted peer data.
Ingestion points: Remote Procedure Calls (RPCs) in scripts/authoritative_looting.gd (request_loot), scripts/state_replication_unreliable.gd (update_player_transform), and scripts/targeted_rpc_relay.gd (talk_to_server) receive data from remote peers.
Boundary markers: The skill explicitly instructs using multiplayer.is_server() checks and strictly synchronizing base types (int, float, vec) rather than objects or resources.
Capability inventory: The skill utilizes the multiplayer API for networking and ResourceLoader for scene instantiation.
Sanitization: The SKILL.md contains a critical security warning to never use get_var(true) on untrusted data to prevent remote code execution and mandates authoritative server-side validation for all game state changes.
[DYNAMIC_EXECUTION]: The script scripts/async_map_loader.gd uses ResourceLoader.load_threaded_request to load map sectors. In the Godot engine, loading resource files (.tscn, .res) can involve the execution of attached scripts. The skill demonstrates best practices by using internal resource paths (res://) for these operations.

godot-genre-battle-royale