The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill explicitly requires the user to 'Run as Administrator' for both installation and operation. This grants the automation framework and any associated Python scripts elevated system permissions, significantly increasing the potential impact of security breaches or unintended system modifications.
[METADATA_POISONING]: The troubleshooting section advises users to 'Disable antivirus software if detection issues arise'. This instruction directly encourages users to compromise their system's security posture, which is a common tactic used to allow malicious software to execute without detection.
[EXTERNAL_DOWNLOADS]: The skill directs users to download releases and core dependencies from several third-party GitHub repositories that are not associated with trusted vendors. These include '1bananachicken/MaaNTE', 'MaaXYZ/MaaFramework', and 'SweetSmellFox/MFAAvalonia'.
[COMMAND_EXECUTION]: The skill leverages the MaaFramework, which allows for 'Custom' actions. These actions are defined in Python scripts (e.g., in the custom/ directory) and can be triggered via JSON pipeline definitions. This architecture allows the execution of arbitrary Python code as part of the automation workflow.
[DATA_EXFILTRATION]: The skill's primary mechanism for automation involves capturing the user's screen ('context.tasker.controller.cached_image'). While intended for game template matching, this capability creates a technical vector for unauthorized screen monitoring or data capture of sensitive information visible on the desktop.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of pipeline JSON files and template images.
Ingestion points: Reads pipeline/*.json, interface.json, and template images from assets/ to determine agent actions.
Boundary markers: Absent. There are no delimiters or instructions to ignore embedded commands within the pipeline data.
Capability inventory: The framework has high-impact capabilities including simulating keyboard/mouse input ('Seize' mode), starting/stopping applications, and executing custom Python scripts.
Sanitization: None. The logic assumes the integrity of the JSON task definitions.

maante-game-automation