The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a mechanism for loading and processing dialogue data from external JSON files, which includes logic strings evaluated at runtime. This creates a surface for indirect prompt injection if dialogue assets are sourced from untrusted third parties.
Ingestion points: DialogueLoader.load_from_json() in references/external-formats.md opens and parses JSON files into game resources.
Capability inventory: The system uses Expression.execute() within the DialogueManager (found in references/dialogue-manager.md) to evaluate logic. This allows execution of GDScript code strings defined in the data.
Sanitization: No sanitization or validation is applied to the condition strings before they are parsed and executed by the engine.
Boundary markers: The dialogue system lacks delimiters or explicit instructions to the engine to ignore embedded commands within the text or condition fields.
[DYNAMIC_EXECUTION]: The GDScript implementation in references/dialogue-manager.md utilizes the Expression class to parse and execute strings at runtime. This allows for dynamic game logic but can be leveraged to execute arbitrary code if the source strings are manipulated.

godot-prompter-dialogue-system