The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements a runtime condition evaluator using Godot's Expression class in references/dialogue-manager.md and references/branching-and-conditions.md. This allows strings stored in the condition property of dialogue data to be executed as code at runtime. While this is a common approach for game logic, it is functionally equivalent to eval() and poses a risk of arbitrary code execution within the game process.
[INDIRECT_PROMPT_INJECTION]: The skill provides logic for ingesting external data that directly influences control flow and execution in references/external-formats.md.
Ingestion points: The DialogueLoader.load_from_json(path) method reads and parses external JSON files into executable dialogue objects.
Boundary markers: Absent. No mechanisms are implemented to separate safe text from executable logic or to prevent the execution of embedded instructions.
Capability inventory: The system allows execution against the GameState singleton or the DialogueManager node, providing access to game state and global methods.
Sanitization: Absent. The implementation does not include validation, escaping, or keyword whitelisting for the condition strings before they are passed to the Expression parser.

dialogue-system