polish

name	polish
description	Apply a fixed quality rubric to existing work and fix every gap: correctness, design, robustness, tests, and public interface. Triggers: polish, tighten, production-grade.

Polish

Use this for completed work that needs a quality pass. Do not use it for metric optimization, red CI, or half-built generations.

Rubric

Score honestly across:

1. Correctness: works in edge cases, not only the happy path.
2. Design: simple structure, justified abstractions, no avoidable complexity.
3. Robustness: errors fail loud and recoverably.
4. Tests: behavior coverage that would catch regressions.
5. Interface: clean names, defaults, docs, CLI/API, and reader path.

Flow

1. Audit the current artifact against the rubric.
2. List concrete issues by severity.
3. Fix issues in priority order.
4. Run meaningful checks.
5. Re-score and repeat, up to five rounds.

Rules

• No score inflation; fine is a 6, not an 8.
• No fluff comments or cosmetic-only changes.
• Broken tests stop the pass until fixed.

Use references/RUBRIC.md for scoring anchors and references/full-reference.md for the old full loop.

Then consider

• evolve for measurable optimization.
• converge if checks are red.