The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill implements a remote execution bridge where the Node.js relay server sends JavaScript code to the Android client. The phone client (scripts/autojs-task-phone-client.js) uses the engines.execScriptFile function to run code downloaded dynamically from the server's /probe/ endpoint. This allows for arbitrary script execution on the mobile device.
[DATA_EXFILTRATION]: Multiple task templates are designed specifically for data retrieval. scripts/tasks/get_clipboard/get_clipboard.js reads the device's clipboard and uploads it to the PC, while scripts/tasks/download_file/download_file.js allows the retrieval of any file from the phone's SD card storage to the local machine.
[COMMAND_EXECUTION]: The skill includes example code (scripts/autojs代码参考例子/侧重UI和canvas和悬浮窗/app启动停止和输入.js) that utilizes the AutoJS shell() function to execute system-level commands, including those requiring root privileges such as am force-stop.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the Android device's screen using OCR (scripts/tasks/ocr/ocr.js) and UI tree inspection (scripts/tasks/inspect_control_by_text/inspect_control_by_text.js).
Ingestion points: Screen content is ingested via captureScreen() and UI selectors in various task scripts.
Boundary markers: There are no explicit boundary markers or instructions to the AI to ignore content found on the screen.
Capability inventory: The skill possesses extensive capabilities including file read/write, shell execution, and remote script execution.
Sanitization: External content is interpolated directly into the agent's context without sanitization, allowing malicious instructions on the phone screen to potentially influence agent behavior.

autojs-task-runner-yashu