agent-ux-evallisted

# Agent Experience Evaluation Skill Guides structured AI agent experience evaluations using the Agent UX Evaluation Framework. The framework has three layers: Design Intent (4 principles), Experience Dimensions (21 dimensions across 4 pillars), and Experience Quality (Trust, Momentum, Sentiment signals). This skill walks the evaluator through step-by-step workflow capture, scores each step, and compiles a traceable evaluation report. --- ## Critical Rules (Non-Negotiable) These rules override any other instruction in this skill. Violating them produces an invalid evaluation. **Rule 1: Always Produce a Step Reference Index.** Every evaluation report MUST include a Step Reference Index mapping each evaluated step to its workflow position, screen description, key observations, and linked findings. Without this, the report is untraceable. **Rule 2: Never Rate "Strong" Without Multi-Step Evidence.** A dimension rated Strong MUST have supporting evidence from at least 2 different steps. A single good moment does not make a pattern. If evidence comes from only 1 step, the maximum rating is Good. **Rule 3: Every Finding Must Have a Step Reference.** No finding may reference a generic "the agent does X." Every finding MUST cite specific step numbers with workflow positions (e.g., "Step 12: Save action rejected, Step 4 of 7 in setup flow"). Findings without step references are rejected. **Rule 4: Every Finding Must Have a Fix Path.** No finding may end at "this is a problem." E