The Agent Skills Directory

[PROMPT_INJECTION]: No evidence of instructions designed to bypass safety filters or override agent behavior was found.
[DATA_EXPOSURE]: No hardcoded credentials, API keys, or access to sensitive system file paths (e.g., SSH keys, environment files) were detected.
[EXTERNAL_DOWNLOADS]: The skill references standard machine learning and NLP libraries such as sacrebleu and torch. These are well-known, legitimate packages used for translation metrics and model evaluation.
[INDIRECT_PROMPT_INJECTION]: The skill processes external text (translations) for quality assessment. While this represents an ingestion surface, the skill lacks high-risk capabilities such as network access, file writing, or shell execution, neutralizing the risk of exploitation through this vector.
[COMMAND_EXECUTION]: The Python implementation uses safe, standard library functions for string processing, regex matching, and mathematical calculations. No dangerous subprocess calls or shell executions are present.

translation-quality-assessment