Stop pasting files into agent prompts (julienreszka.com)

A new approach has emerged for optimizing interactions with AI agents, particularly in handling structured files. Instead of feeding raw files directly to agents, the recommendation is to develop an audit script that classifies and scores issues within the data, outputting a concise CSV. This method is significant for the AI/ML community as it allows agents to process information more efficiently, effectively replacing complex source content with streamlined issue lists. By leveraging the familiarity of CSV formats from LLM training data, agents can understand the input without needing additional explanations. The audit script utilizes six key rules for effective problem identification: issues are classified by type, scored by severity, deduplicated, and accompanied by context snippets. Additionally, using emojis for severity ratings enhances human readability while keeping the output succinct. This process mirrors test-driven development, wherein the audit script acts as a compiler for data, enabling systematic issue resolution while maintaining observability through a decreasing row count in the CSV. This advanced method not only makes working with structured files more efficient but also allows agents to better utilize their built-in capabilities for regex and problem-solving.