The Agent Skills Directory

[SAFE]: The skill provides standard architectural patterns for game persistence without any malicious behavior. It emphasizes security-conscious practices like atomic writes to prevent file corruption and client-side validation to maintain data integrity.
[INDIRECT_PROMPT_INJECTION]: The skill outlines patterns for loading data from external files (game saves), which is a potential ingestion surface for untrusted data. However, it explicitly mitigates this through mandatory validation steps.
Ingestion points: load_save function patterns in SKILL.md and references/versioning-and-migration.md read data from local disk files.
Boundary markers: The instructions recommend structured data formats (JSON/Binary) to clearly delimit data fields.
Capability inventory: Includes file system write and rename capabilities through Godot's FileAccess and DirAccess classes.
Sanitization: The skill mandates a specific "Load-time validation checklist" including type verification, range checking (e.g., hp ≥ 0), and ensuring enum/ID values exist in the current game content.
[COMMAND_EXECUTION]: The skill includes code snippets for file system operations necessary for save state management.
Evidence: Uses FileAccess.open(tmp, FileAccess.WRITE) and DirAccess.rename_absolute(tmp, path) in SKILL.md to implement atomic, crash-safe file persistence.

save-systems