The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external narrative data, which serves as a potential surface for indirect prompt injection if the source data is untrusted.
Ingestion points: scripts/dialogue_engine.gd includes a load_dialogue_graph function that parses external JSON files. scripts/dialogue_manager_singleton.gd processes DialogueResource objects.
Capability inventory: The system can emit gameplay signals via scripts/dialogue_event_bridge.gd, record analytics via scripts/dialogue_stat_logger.gd using OS.add_logger, and manipulate UI elements.
Boundary markers: No specific delimiters or "ignore instructions" warnings are implemented for the dialogue text processing.
Sanitization: The skill relies on Godot's built-in JSON and Resource parsing, with no additional narrative-layer sanitization for embedded instructions.
[DYNAMIC_EXECUTION]: The skill architecture includes placeholders for dynamic logic evaluation.
Evidence: scripts/dialogue_option_data.gd defines a condition_script property intended for GDScript snippets. scripts/branching_condition_validator.gd notes the intent for "Dynamic evaluation logic".
Context: No actual execution mechanism (such as Expression.execute() or eval) is implemented in the provided code; the fields currently act as metadata.
[COMMAND_EXECUTION]: The skill uses Godot's OS and FileAccess singletons for legitimate game development purposes.
Evidence: scripts/dialogue_stat_logger.gd uses OS.add_logger() to register a custom analytics logger. scripts/dialogue_engine.gd uses FileAccess.open() to load dialogue data from the local file system.

godot-dialogue-system