The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs shell utilities such as unzip, zip, and rm -rf using interpolated variables for directory names and document paths (e.g., {임시디렉토리}, {서식}.hwpx). This creates a command injection vulnerability if a malicious filename or path containing shell metacharacters is processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill instructs the installation of the pyhwpx Python package via pip, which introduces an external dependency from a non-pre-approved source.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it processes untrusted content from documents and user input.
Ingestion points: Hangul documents (.hwp, .hwpx) and Markdown input.
Boundary markers: None implemented to delimit or isolate external data.
Capability inventory: Shell execution, file system writes, and external package installation.
Sanitization: Limited to basic XML character escaping (e.g., &, <, >).

hwp-fill