ux-ai-first-design

What this skill does

<!-- VERSION: 1.1.0 | DATE: 2026-03-04 | SOURCE: skills/user-experience/SKILL.md | PARENT: /user-experience skill | REVISION: iter5 — version comment alignment; iter4: applied iter3 defect fixes (Shneiderman DOI correction to IJHCI 10.1080/10447318.2020.1741118, Stage 5 operational criterion, classification tie-breaker rules), added WSM file path pointer to wave-progression.md, explicit default-case rules for classification algorithms; iter3: version 1.0.0→1.1.0, error-rate threshold brackets, progressive disclosure calibration footnote, confidence:0.5 explanatory comment -->

# AI-First Design Sub-Skill

> **Version:** 1.1.0
> **Framework:** Jerry User-Experience -- AI-First Interaction Design
> **Constitutional Compliance:** Jerry Constitution v1.0
> **Parent Skill:** `/user-experience` (`skills/user-experience/SKILL.md`)
> **Wave:** 5 (Process Intensives) -- CONDITIONAL
> **Project:** PROJ-022 User Experience Skill | GitHub Issue [#138](https://github.com/geekatron/jerry/issues/138)

## Document Sections

| Section | Purpose |
|---------|---------|
| [Document Audience](#document-audience-triple-lens) | Triple-Lens audience guide |
| [Purpose](#purpose) | Sub-skill overview and key capabilities |
| [When to Use This Sub-Skill](#when-to-use-this-sub-skill) | Activation triggers, scope boundaries, and CONDITIONAL activation note |
| [Available Agents](#available-agents) | Single agent with role, model, and output location |
| [P-003 Compliance](#p-003-compliance) | Worker agent hierarchy position |
| [Invoking the Agent](#invoking-the-agent) | Invocation via ux-orchestrator with conditional activation check |
| [Methodology](#methodology) | AI-first interaction design process with trust-risk/error-risk classification |
| [Output Specification](#output-specification) | Output location, L0/L1/L2 structure, required sections |
| [Routing](#routing) | Keywords and lifecycle-stage routing integration |
| [Cross-Framework Integration](#cross-framework-integration) | Handoff from heuristic eval, handoff to inclusive design for AI accessibility |
| [Synthesis Hypothesis Confidence](#synthesis-hypothesis-confidence) | Confidence classifications for AI-First Design outputs |
| [Quality Gate Integration](#quality-gate-integration) | S-014 scoring and H-13 threshold enforcement |
| [Degraded Mode Behavior](#degraded-mode-behavior) | Operation without Figma MCP |
| [Wave Architecture](#wave-architecture) | Wave 5 entry criteria, CONDITIONAL activation, bypass |
| [Constitutional Compliance](#constitutional-compliance) | Governing principles including P-022 AI pattern staleness disclosure |
| [Registration](#registration) | H-26 parent-routed registration model and AGENTS.md confirmation |
| [Deployment Status](#deployment-status) | Wave 5 stub agent status and implementation timeline |
| [Quick Reference](#quick-reference) | Common workflows and agent selection hints |
| [References](#references) | Full repo-relative paths, requirements traceability, external citations |

## Document Audience (Triple-Lens)

This SKILL.md serves multiple audiences:

| Level | Audience | Sections to Focus On |
|-------|----------|---------------------|
| **L0 (Stakeholder)** | Product managers, designers | [Purpose](#purpose), [When to Use This Sub-Skill](#when-to-use-this-sub-skill), [Quick Reference](#quick-reference) |
| **L1 (Developer)** | Engineers invoking the agent | [Invoking the Agent](#invoking-the-agent), [Methodology](#methodology), [Output Specification](#output-specification) |
| **L2 (Architect)** | Workflow designers, skill maintainers | [Cross-Framework Integration](#cross-framework-integration), [Synthesis Hypothesis Confidence](#synthesis-hypothesis-confidence), [Degraded Mode Behavior](#degraded-mode-behavior), [Wave Architecture](#wave-architecture) |

---

## Purpose

The AI-First Design sub-skill provides structured AI interaction design guidance using trust-calibrated pattern selection based on Yang et al.'s (2020) two primary failure modes in human-AI interaction: trust miscalibration and error cost mismanagement. It targets tiny teams (1-5 people) building AI-powered features who need a systematic framework to design interactions that appropriately calibrate user trust, manage AI error consequences, and progressively disclose AI capabilities.

This sub-skill is part of Wave 5 (Process Intensives) and is **CONDITIONAL** -- it only activates when WSM >= 7.80 AND enabler research (FEAT-020) is complete. When the condition is not met, the `ux-orchestrator` routes to `/ux-heuristic-eval` with PAIR protocol (AI-specific heuristic supplement) as an interim alternative.

The sub-skill bridges behavioral diagnosis (Wave 4) and full AI product maturity by providing interaction design guidance that accounts for the unique challenges of AI-powered interfaces: non-deterministic outputs, confidence-dependent reliability, and the need for graceful degradation when AI predictions are wrong.

### Key Capabilities

- **Trust-Risk Classification** -- Assesses how much users should trust AI outputs across a three-level scale (HIGH: AI makes decisions autonomously, MEDIUM: AI recommends with human decision, LOW: AI assists with human control), based on Yang et al.'s (2020) trust miscalibration failure mode
- **Error-Risk Classification** -- Evaluates the cost of AI errors across a three-level scale (HIGH: irreversible or safety-critical, MEDIUM: costly but recoverable, LOW: negligible impact), based on Yang et al.'s (2020) error cost mismanagement failure mode
- **Interaction Pattern Selection** -- Selects interaction patterns from the trust-risk x error-risk matrix, mapping each combination to an appropriate human-AI collaboration level (full human oversight, AI-as-advisor, human-in-the-loop, AI autonomous with monitoring, AI autonomous)
- **Feedback Loop Design** -- Designs how the AI system communicates uncertainty, explains decisions, handles errors gracefully, and enables user correction, following Amershi et al.'s (2019) 18 guidelines for human-AI interaction
- **Progressive Disclosure of AI Capabilities** -- Designs graduated onboarding experiences that build user trust over time, starting with low-stakes AI features and progressively revealing higher-autonomy capabilities as trust is established (Shneiderman, 2020)
- **AI Transparency Assessment** -- Evaluates current AI interface transparency against the Google PAIR (2019) guidebook patterns, identifying where explainability gaps erode user trust or where excessive explanation creates cognitive overload

---

## When to Use This Sub-Skill

**CONDITIONAL ACTIVATION:** This sub-skill only activates when WSM >= 7.80 AND enabler research (FEAT-020) is complete. If either condition is unmet, the `ux-orchestrator` routes to `/ux-heuristic-eval` with PAIR protocol (AI-specific heuristic supplement) as an interim alternative.

Activate when:

- Designing AI-powered features and needing to determine the appropriate level of user trust and human oversight
- Classifying AI interaction patterns by trust-risk and error-risk to select the right human-AI collaboration model
- Designing feedback loops for AI systems that communicate uncertainty, explain decisions, and handle errors gracefully
- Planning progressive disclosure strategies for AI capabilities that build user trust incrementally
- Auditing existing AI interfaces for trust calibration issues, transparency gaps, or inappropriate autonomy levels
- Designing conversational UX for LLM-powered features (chatbots, AI assistants, agentic interfaces)
- Evaluating whether an AI feature needs human-in-the-loop oversight vs. autonomous operation
- Designing error recovery flows specific to AI prediction failures or hallucination scenarios

Do NOT use for:

- Evaluating an existing non-AI interface against usability heuristics -- use `/ux-heuristic-eval` (Nielsen's 10) instead. For AI interfaces where the CONDITIONAL activation is not met, use `/ux-heuristic-eval` with PAIR prot