The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's code templates in references/code-templates.md recommend the use of standard, well-known libraries such as anthropic, pandas, and scikit-learn for Python, and httr2, jsonlite, and caret for R. These are standard dependencies for data science and LLM integration.
[DATA_EXFILTRATION]: The provided batch processing templates facilitate sending text data to api.anthropic.com. As this is a well-known technology service and the primary function of the skill is to perform LLM-based classification, this network activity is documented neutrally as a core feature.
[COMMAND_EXECUTION]: The skill provides Python and R scripts intended for the user to execute in their local environment. These scripts perform routine tasks such as reading local CSV files and making API requests. There is no evidence of the agent attempting to execute unauthorized commands or scripts.
[PROMPT_INJECTION]: The skill is designed to process untrusted text data for classification purposes. While the templates do not explicitly include complex boundary markers to prevent indirect prompt injection, the skill's workflow (Phases 0-6) is built around a 'human-in-the-loop' model, where the user (domain expert) reviews results, examines misclassifications, and validates the process through 'memos'. This iterative, reflective approach significantly mitigates the risks associated with processing external content.

prompt-optimizer