The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements a privilege escalation mechanism using Windows system calls.
Evidence: The request_admin function uses ctypes.windll.shell32.ShellExecuteW(None, "runas", sys.executable, ...) to restart the current script with administrative privileges.
[REMOTE_CODE_EXECUTION]: The skill performs direct memory manipulation and injection into external processes.
Evidence: The use of the pymem library (e.g., self.pm.write_int(address, value)) enables the skill to modify the execution state and data of other running applications, such as DELTARUNE.exe.
[PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection through external configuration files and contains deceptive metadata.
Evidence:
Ingestion points: The load_config function reads data from an external config.json file in SKILL.md.
Boundary markers: No delimiters or safety instructions are used to separate the external configuration data from the trainer's execution logic.
Capability inventory: The skill possesses significant capabilities including process memory writing and administrator privilege escalation.
Sanitization: The skill lacks validation or sanitization for the memory addresses and offsets loaded from the JSON file.
Deceptive Content: The skill includes a disclaimer claiming the repository is malware and references a non-existent game chapter, which serves as a form of metadata poisoning or deceptive instruction.

deltarune-game-trainer-client