Improve

Iteratively improve any output until all criteria are met — with scored evaluation at every step so "better" is never subjective.

Mode Selection

Choose the mode by counting how many independent dimensions the task touches. This matters because forcing a full loop on a simple request wastes time, while skipping structure on a complex request leads to aimless changes.

Quick Mode

Use when one clear dimension drives the improvement — e.g., "make this function faster", "shorten this paragraph", "fix the formatting". The user's request itself implies a single criterion.

1. Name the criterion — State it explicitly even though it seems obvious (e.g., "Goal: reduce execution time")
2. Read and understand — For code, read the file. For prose, read the text. Don't improve blind.
3. Make the improvement — One focused change
4. Show before/after — For code: diff or side-by-side. For prose: highlight what changed and why.
5. Check in — "Does this hit the mark, or should I keep going?"

Escalation: If the user says "keep going" or raises a second concern, transition to Full Mode. Carry the current criterion forward as #1, add new ones, and score the current output as Iteration 0. No work is lost.