The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and rewrite untrusted user-provided Korean text. It mitigates the risk of indirect prompt injection by enforcing '4-item ironclad rules' for meaning invariance and a '6-item self-verification' checklist that ensures facts, proper nouns, and citations remain unchanged. Capability inventory includes local workspace file operations and execution of bundled Python scripts for linguistic metrics. Sanitization is handled through explicit 'Do-NOT' lists that exclude sensitive data types from processing.
[SAFE]: The skill executes local Python scripts (metrics.py, metrics_v2.py, humanize_html.py) to perform its primary analysis tasks. These scripts rely exclusively on Python standard libraries and do not perform network requests, access sensitive system paths, or use dangerous functions like eval on user input.
[SAFE]: All external references, including GitHub links and academic identifiers (arXiv, DOI), are legitimate and directly related to the skill's functional domain of Korean translation studies and AI text detection.

korean-humanize