The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts (metrics.py and metrics_v2.py) to analyze text metrics. These scripts use standard Python libraries, do not perform network operations, and are restricted to reading/writing files within the designated local workspace.
[DATA_EXFILTRATION]: No evidence of unauthorized data access or network-based exfiltration was found. The skill processes user-supplied text locally and stores intermediate results in a structured _workspace directory.
[PROMPT_INJECTION]: The skill's instructions are focused on its primary purpose of Korean text refinement. It includes several 'Prime Directives' and 'Self-Verification' steps designed to ensure the agent maintains the original meaning and stays within specific stylistic boundaries, which act as safeguards against unintended behavior.
[REMOTE_CODE_EXECUTION]: The skill does not download or execute code from external sources. All processing is handled by provided local scripts and the agent's internal logic.
[INDIRECT_PROMPT_INJECTION]: As the skill is designed to process untrusted user-supplied text for rewriting, it naturally has an indirect prompt injection surface. However, the skill implements strict fidelity checks (100% preservation of numbers, dates, and names) and change-rate limits (thresholds at 30% and 50%) to mitigate the risk of the input text successfully hijacking the agent's instructions.

humanize-korean