Prompt Optimizer

Takes a rigorous engineering approach to prompt writing: you define success criteria and eval cases first, then iterate systematically. It walks you through capturing requirements, inventorying external context by file path, choosing model-specific strategies (OpenAI, Claude, Gemini), and running comparison loops with concrete failure clustering. The methodology is opinionated about layer separation, keeping stable policy in system prompts and dynamic context in user payloads, and cutting anything that doesn't improve evals. Useful when you're debugging flaky agent behavior, porting prompts between models, or just tired of prompt rewrites that feel like superstition. Includes reference docs on core patterns and model family quirks.