The Agent Skills Directory

[SAFE]: The skill provides a robust framework for text transformation across nine languages (EN, RU, UK, DE, FR, ES, PT, IT, PL). All logic is transparently documented in the SKILL.md and shared/ directory.\n- [DATA_EXFILTRATION]: The skill includes optional benchmarking tools in the scripts/ directory that interface with the well-known ZeroGPT API. This is a documented feature for quality evaluation and requires a user-provided API key; it is not a silent exfiltration mechanism.\n- [COMMAND_EXECUTION]: Local automation scripts (PowerShell and Bash) are provided for integrity checking (validate.sh) and benchmarking (run-benchmark.ps1). These scripts use standard system commands and do not exhibit malicious behavior or unauthorized privilege escalation.\n- [PROMPT_INJECTION]: The skill contains logic to detect and remove common AI-generated phrases such as 'Ignore previous instructions'. These are used as detection markers for the humanization process and do not represent attempts to inject malicious instructions into the host agent.\n- [SAFE]: Analysis of all 10 threat categories confirms that the skill is safe for use. External dependencies and API references are limited to well-known services (ZeroGPT) and are consistent with the skill's stated purpose of AI detection and text humanization.

human-ai