creative-eye_skill

Overview

This skill gives AI agents practical creative judgment and design taste to produce and evaluate visual content before publishing. It supplies a step-by-step framework, scorecards, evaluation prompts, and reference-library patterns so agents can learn, compare, create intentionally, and gate quality. Use it to raise creative standards, avoid common aesthetic failures, and enforce brand consistency.

How this skill works

The skill guides agents through STUDY → COMPARE → CREATE → EVALUATE. Agents build a reference library and JSON style profiles, select 3–5 references before generating, justify every design decision, then run automated vision-model critiques and scorecards. Low scores trigger targeted fixes or kill decisions, with a maximum of three iterative attempts before escalation.

When to use it

• Creating any visual content: merch, social posts, ads, product photography, brand materials
• Evaluating creative quality before publishing or running campaigns
• Building a reference library or extracting JSON style profiles
• Setting up pre-publish creative quality gates and brand guardrails
• Training an agent to develop aesthetic judgment and self-refinement

Best practices

• Study one great piece daily and log concrete, technical observations into a study file
• Always pick 3–5 format-specific references and derive explicit constraints from them
• Require a rationale for every design choice: typography, color, layout, image treatment
• Run the 5-point creative scorecard and 10-point pre-publish checklist on every asset
• Use a vision model to critique images against references, then apply targeted fixes only (max 3 iterations)
• Document failures with root-cause analysis and anti-pattern rules to prevent repeat mistakes

Example use cases

• Pre-publish review for a social ad: get scroll-stop power, authenticity, and brand-voice feedback
• Merch design QA: evaluate typography, wearability, and production quality before sample runs
• Product photography audit: check lighting, composition, technical quality, and brand story
• Automated brand guardrails: block assets that fail hard-stop visual rules before posting
• Training an agent: feed reference JSON style profiles so generated images match a target aesthetic