The Agent Skills Directory

[COMMAND_EXECUTION]: The documentation and installation guides (README.md, install/claude-code.md, install/codex.md) provide standard shell commands for the user to execute for installation and maintenance, such as git clone, mkdir, cp, and ln -s. These are instructions for the user, not automated triggers for the AI.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to download the repository from a trusted vendor source (github.com/tentenco/shuorenhua-zh-tw.git) during installation. This is a standard procedure for installing skills in environments like Claude Code.
[DATA_EXFILTRATION]: No commands or patterns were found that attempt to read sensitive files (e.g., credentials, SSH keys) or send data to unauthorized external endpoints. The automation scripts in the automation/ directory output data to local files within a .gitignoreed directory.
[PROMPT_INJECTION]: The skill contains no instructions designed to bypass agent safety filters or override system prompts. On the contrary, it includes extensive rules in references/protected-spans.md and references/severity.md to prevent the AI from modifying critical information like commands, code, and specific technical terms.
[SAFE]: The repository demonstrates high security awareness by providing a 'Protected Spans' pre-check and an 'Audit-only' mode for processing citations, ensuring that the AI does not fabricate facts or accidentally alter executable code snippets while performing text humanization.

shuorenhua