What problem does it solve? Choosing a writing style for a prompt (caveman, terse, formal, bullet) is usually guesswork, and shorter prompts trade quality for token savings in unknown ways. This Skill measures that tradeoff empirically so you can pick a style based on data instead of intuition. ## Core Features & Use Cases - Style rewriting: Rewrites a base prompt into each style defined in src/styles/index.ts (baseline, caveman, terse, formal, bullet) via the patterns/restyle.js workflow. - Adaptive sweep: Runs all style variants over shared eval cases with patterns/eval-sweep.js, using adaptive seed batches up to n_max with a variance ceiling. - Cost-join and Pareto frontier: Joins quality scores with token costs via dist/styles/cost-join.js and computes a Pareto frontier and quality_per_cost winner through dist/compose-eval/cli.js. - Use Case: You suspect a terse version of your classify-text prompt would cut token spend. Run this Skill to get a report showing each style's quality mean, token cost, tokens saved versus baseline, and which styles dominate baseline. ## Quick Start Ask the agent to test writing styles for the classify-text skill using src/skills/classify-text/variants/v1-baseline.md as the base prompt and show the Pareto frontier.
