The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bundled Python utility (scripts/prepare_monolith_input.py) via shell commands to calculate linguistic metrics from user input. This process is limited to local script execution for data analysis purposes.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided Korean text, which constitutes an attack surface for indirect prompt injection.
Ingestion points: User-provided Korean text or file paths are ingested via the SKILL.md workflow.
Boundary markers: The pre-processing script implements explicit [원문 시작] and [원문 끝] markers to separate user data from instructions in the generated metrics block.
Capability inventory: The skill can perform local file system writes to a workspace directory and execute bundled scripts.
Sanitization: The text is treated as data for linguistic evaluation, and the skill includes strict instructions to preserve meaning and avoid arbitrary additions.
[DYNAMIC_EXECUTION]: The prepare_monolith_input.py script performs dynamic loading of its own local modules by modifying sys.path at runtime. This behavior is used to maintain compatibility between metric versions and does not involve executing external or untrusted code.

humanize-korean