refinery

Score, diagnose, rewrite, and re-score artifacts until quality thresholds are met.

3|1|Updated Feb 9, 2026
One-click install
npx skills add https://github.com/vxcozy/architect-system --skill refinery
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: refinery
Source: https://github.com/vxcozy/architect-system/tree/main/.claude/skills/refinery
Command: npx skills add https://github.com/vxcozy/architect-system --skill refinery

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Convergence loop that turns imperfect work into high-quality outcomes by scoring, diagnosing, rewriting, and re-scoring until quality thresholds are met. It reads analyst reviews and prior refinery history to guide targeted improvements, ensuring deterministic progress.

Core Features & Use Cases

  • Reads the latest analyst reviews from system/reviews/* and prior refinery logs to generate concrete improvement targets.
  • Scores artifacts across defined dimensions, diagnoses root causes, applies rewrites, and re-scores to verify progress.
  • Use Case: after a REVISE verdict or when output quality is below 8/10, run the refinery loop to converge toward the threshold.

Quick Start

Run the refinery loop on the artifact that needs improvement to iteratively converge its quality.

Frequently Asked Questions about refinery

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I iteratively refine work to meet a quality threshold?

Iterative refinement automates scoring, diagnosing, rewriting, and re-scoring artifacts until quality thresholds are met. It reads analyst reviews and prior refinery logs to generate concrete improvement targets and ensure deterministic progress.

When should I run an automated rewrite and review loop?

Run an automated rewrite and review loop after a REVISE verdict from an analyst or when output quality falls below the 8/10 target. This drives convergence by applying targeted improvements based on prior diagnostic history.

How does artifact scoring and diagnosis work for quality assurance?

Artifact scoring evaluates work across defined dimensions, diagnoses root causes of low scores, and applies rewrites. It verifies progress by re-scoring the updated artifact to ensure quality metrics improve toward the target threshold.

Do I need prior analyst reviews to run the refinement loop?

Prior analyst reviews from system/reviews are required to guide targeted improvements during the refinement loop. The process also reads system/refinery-log.md to track iteration history and ensure deterministic convergence toward the quality threshold.

What's the best way to reduce friction in iterative quality assurance?

To reduce friction in iterative quality assurance, automate the convergence loop of scoring, diagnosing, and rewriting. This ensures deterministic progress toward the 8/10 target without manual intervention for each revision cycle.

Why does my output quality fail to converge after multiple rewrites?

Output quality may fail to converge if root causes are not properly diagnosed before rewriting. The refinery loop reads prior history and analyst reviews to apply targeted improvements, ensuring that re-scoring verifies actual progress toward the threshold.