The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a system/shell action capability specifically for the connected Android device. This allows for executing shell commands, potentially with root privileges (is_root: true), to perform system-level tasks such as launching specific activities. This functionality is transparently documented and scoped to the target device provided by the user.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to an Android device at a user-specified base-url. Additionally, the generated child skills are designed to communicate with a CONTROL_API_URL environment variable. All network activity is directed toward infrastructure provided or configured by the user.
[DATA_EXFILTRATION]: While the skill reads sensitive UI data (accessibility dumps) from the connected Android device, this data is stored locally in the authoring/ workspace for the purpose of script generation. No evidence of unauthorized data transmission to remote servers was found.
[SAFE]: The skill's architecture is robust, utilizing sanitization methods like jsonStringEscape and decodeXmlEntities when processing device data. The code generation for child skills follows a static templating approach, and no obfuscation, prompt injection, or persistence mechanisms were detected in the scripts or documentation.

workflow-skill-creator