The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/dynamic_script_attachment.gd implements a runtime script attachment mechanism using node.set_script(load(script_path)). Additionally, references/scene-patterns-deep.md documents the use of ProjectSettings.load_resource_pack(pck_path) for mounting external resource packs. These patterns allow for the execution of code from computed or external paths, which is a standard requirement for modding but constitutes dynamic loading.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface where untrusted data could influence agent actions.
Ingestion points: Methods such as apply_script_to_node(node, script_path) in scripts/dynamic_script_attachment.gd, load_scene(path) in scripts/async_scene_manager.gd, and open_menu(path) in scripts/additive_ui_layering.gd accept path strings that are processed by the agent.
Boundary markers: None identified; there are no explicit delimiters or instructions to ignore embedded content in loaded resources.
Capability inventory: The skill uses load().instantiate() and node.set_script(), which can trigger the execution of logic attached to Godot scenes or script resources.
Sanitization: The provided scripts do not implement validation or sanitization for the path or script_path arguments before passing them to the load() function.

godot-scene-management