rulph

What problem does it solve?

It prevents quality drift by turning subjective “looks good” reviews into an iterative rubric loop that can repeatedly score and improve an artifact until it passes a target threshold and per-criterion floor.

Core Features & Use Cases

• Interactive rubric building: You define criteria first via structured prompts, then lock a weighted scoring contract.
• Multi-model parallel evaluation: The skill evaluates the same artifact concurrently with available models (Codex, Gemini, and Claude via subagent) and aggregates results.
• Criterion-by-criterion autonomous improvement: It repeatedly improves only the single lowest-impact or floor-violating criterion, re-scores, and continues until success or a circuit breaker stops the loop.

Quick Start

Use the rulph skill to iteratively grade and improve an artifact you care about by asking it to evaluate a file (or text) against criteria you provide, then let it loop until the required threshold and floors are met.