The Agent Skills Directory

[DATA_EXFILTRATION] (INFO): The code initiates network requests to external LLM providers (OpenAI, Anthropic) via the DSPy library for model evaluation. This is standard and expected behavior for this framework.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill demonstrates processing external data (ticket_text) through a language model classifier. This creates a potential surface for indirect prompt injection where malicious input could influence the model's logic. However, as the example focuses on classification and logging with no high-privilege side effects, the risk is negligible.
[EXTERNAL_DOWNLOADS] (LOW): The code depends on the dspy package, which is a standard library for programmatic LLM usage.

ai-tracking-experiments