The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file scripts/save_migration_manager.gd contains a hardcoded encryption key SECRET_XP_KEY_2026 used for AES-256 encryption. Hardcoding cryptographic keys allows anyone with access to the source code to decrypt user save data or forge malicious save files.
[DYNAMIC_EXECUTION]: In SKILL.md, 'Pattern 2: Binary Save System' demonstrates the use of get_var(true). The true parameter enables 'allow_objects', which permits the instantiation of arbitrary Godot objects during deserialization. This is a classic insecure deserialization vulnerability.
[REMOTE_CODE_EXECUTION]: The use of get_var(true) on save files—which are often shared or downloaded by players—allows an attacker to craft a malicious save file that executes arbitrary code or performs unauthorized actions upon loading.
[INDIRECT_PROMPT_INJECTION]: The skill implements a significant attack surface for indirect injection via processed data files.
Ingestion points: load_game(), load_game_binary(), and load_encrypted() functions in SKILL.md and referenced scripts ingest data from external files in the user:// directory.
Boundary markers: Absent. The skill lacks specific delimiters or 'ignore' instructions for the data being parsed into the game state.
Capability inventory: The skill utilizes FileAccess for disk writes and get_var(true) for object instantiation, providing a high-impact capability set if the input data is compromised.
Sanitization: While SKILL.md mentions validation using data.get(), the actual implementations in the scripts lack strict schema validation or type checking for complex nested data.

godot-save-load-systems