Agent Skills: Software UX Research Skill — Quick Reference

Software UX Research Skill — Quick Reference

Use this skill to identify problems/opportunities and de-risk decisions. Use software-ui-ux-design to implement UI patterns, component changes, and design system updates.

Mar 2026 Baselines (Core)

• Human-centred design: Iterative design + evaluation grounded in evidence (ISO 9241-210:2019) https://www.iso.org/standard/77520.html
• Usability definition: Effectiveness, efficiency, satisfaction in context (ISO 9241-11:2018) https://www.iso.org/standard/63500.html
• Accessibility baseline: WCAG 2.2 is a W3C Recommendation (12 Dec 2024) https://www.w3.org/TR/WCAG22/
• WCAG 3.0 preview: Working Draft published Sep 2025; introduces Bronze/Silver/Gold conformance tiers and enhanced cognitive accessibility; not expected before 2028-2030 https://www.w3.org/WAI/standards-guidelines/wcag/wcag3-intro/
• EU shipping note: European Accessibility Act applies to covered products/services after 28 Jun 2025 (Directive (EU) 2019/882) https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32019L0882

When to Use This Skill

• Discovery: user needs, JTBD, opportunity sizing, mental models.
• Validation: concepts, prototypes, onboarding/first-run success.
• Evaluative: usability tests, heuristic evaluation, cognitive walkthroughs.
• Quant/behavioral: funnels, cohorts, instrumentation gaps, guardrails.
• Research Ops: intake, prioritization, repository/taxonomy, consent/PII handling.
• Demographic research: Age-diverse, cultural, accessibility participant recruitment.
• A/B testing: Experiment design, sample size, analysis, pitfalls.
• Non-technical user research: Digital literacy assessment, simplified-flow validation, low-tech-confidence usability testing.

When NOT to Use This Skill

• UI implementation → Use software-ui-ux-design for components, patterns, code
• Analytics instrumentation → Use marketing-product-analytics for tracking plans and qa-observability for implementation patterns
• Accessibility compliance audit → Use accessibility-specific checklists (WCAG conformance)
• Marketing research → Use marketing-social-media or related marketing skills
• A/B test platform setup → Use experimentation platforms (Statsig, GrowthBook, LaunchDarkly)

Operating Mode (Core)

If inputs are missing, ask for:

• Decision to unblock (what will change based on this research).
• Target roles/segments and top tasks.
• Platforms and contexts (web/mobile/desktop; remote/on-site; assisted tech).
• Existing evidence (analytics, tickets, reviews, recordings, prior studies).
• Constraints (timeline, recruitment access, compliance, budget).

Default outputs (pick what the user asked for):

• Research plan + output contract (prefer ../software-clean-code-standard/assets/checklists/ux-research-plan-template.md; use assets/research-plan-template.md for skill-specific detail)
• Study protocol (tasks/script + success metrics + recruitment plan)
• Findings report (issues + severity + evidence + recommendations + confidence)
• Decision brief (options + tradeoffs + recommendation + measurement plan)

Required Output Sections

Every research output — plans, protocols, evaluations, reports — must include these sections. They represent the skill's core value beyond standard UX knowledge: governance, confidence calibration, and ethical research practice.

1. Method Justification: Name the chosen method AND explain why alternatives were rejected. Do not just describe the method; explain why it was selected over at least 2 alternatives given the specific context (stage, timeline, sample, question type).

2. Confidence & Triangulation Assessment: Tag every recommendation or finding with a confidence level:

   | Confidence | Evidence requirement | Use for | |------------|----------------------|---------| | High | Multiple methods or sources agree | High-impact decisions | | Medium | Strong signal from one method + supporting indicators | Prioritization | | Low | Single source / small sample | Exploratory hypotheses only |

3. Consent & Data Handling: Include a PII/consent section in every plan or protocol. Research that involves participants requires explicit attention to:

   • Minimum PII collection
   • Identity stored separately from study data
   • Name/email redaction before broad sharing
   • Recording access restricted to need-to-know
   • Consent, purpose, retention, and opt-out documented

4. Decision Framework: For evaluations and analysis outputs, provide a structured decision table with options, confidence levels, timelines, and risks — not just a single recommendation.

5. Pre-Decision Checklist: For experiment evaluations (A/B tests, etc.), include a verification checklist of confounds and data quality checks to complete before any ship/kill decision.

Method Chooser (Core)

Decision Tree (Fast)

What do you need?
  ├─ WHY / needs / context → interviews, contextual inquiry, diary
  ├─ HOW / usability → moderated usability test, cognitive walkthrough, heuristic eval
  ├─ WHAT / scale → analytics/logs + targeted qual follow-ups
  └─ WHICH / causal → experiments (if feasible) or preference tests

When selecting a method, always justify the choice by explaining why 2+ alternatives were rejected given the user's specific context. This is a key differentiator — generic "we'll do interviews" without justification is insufficient.

Research by Product Stage

Stage Framework (What to Do When)

| Stage | Decisions | Primary Methods | Secondary Methods | Output | |-------|-----------|-----------------|-------------------|--------| | Discovery | What to build and for whom | Interviews, field/diary, journey mapping | Competitive analysis, feedback mining | Opportunity brief + JTBD + Forces of Progress | | Concept/MVP | Does the concept work? | Concept test, prototype usability | First-click/tree test | MVP scope + onboarding plan | | Launch | Is it usable + accessible? | Usability testing, accessibility review | Heuristic eval, session replay | Launch blockers + fixes | | Growth | What drives adoption/value? | Segmented analytics + qual follow-ups | Churn interviews, surveys | Retention drivers + friction | | Maturity | What to optimize/deprecate? | Experiments, longitudinal tracking | Unmoderated tests | Incremental roadmap |

Discovery Outputs: Beyond Basic JTBD

Discovery research should produce more than job statements. Include:

• Forces of Progress diagram: Map the four forces acting on switching behavior — Push (current pain), Pull (new solution appeal), Anxiety (fear of change), Habit (inertia). These forces explain why users do or don't adopt, which directly informs positioning and onboarding.
• Pain Point Severity Matrix: Score each pain point by Frequency × Impact × Breadth to prioritize objectively. A pain that affects 3 roles weekly outranks one that affects 1 role monthly, even if the single-role pain feels more dramatic in interviews.

Research for Complex Systems (Workflows, Admin, Regulated)

Complexity Indicators

| Indicator | Example | Research Implication | |-----------|---------|----------------------| | Multi-step workflows | Draft → approve → publish | Task analysis + state mapping | | Multi-role permissions | Admin vs editor vs viewer | Test each role + transitions | | Data dependencies | Requires integrations/sync | Error-path + recovery testing | | High stakes | Finance, healthcare | Safety checks + confirmations | | Expert users | Dev tools, analytics | Recruit real experts (not proxies) |

Evaluation Methods (Core)

• Contextual inquiry: observe real work and constraints.
• Task analysis: map goals → steps → failure points.
• Cognitive walkthrough: evaluate learnability and signifiers.
• Error-path testing: timeouts, offline, partial data, permission loss, retries.
• Multi-role walkthrough: simulate handoffs (creator → reviewer → admin).

Multi-Role Coverage Checklist

• [ ] Role-permission matrix documented.
• [ ] “No access” UX defined (request path, least-privilege defaults).
• [ ] Cross-role handoffs tested (notifications, state changes, audit history).
• [ ] Error recovery tested for each role (retry, undo, escalation).

Research Ops & Governance (Core)

Intake (Make Requests Comparable)

Minimum required fields:

• Decision to unblock and deadline.
• Research questions (primary + secondary).
• Target users/segments and recruitment constraints.
• Existing evidence and links.
• Deliverable format + audience.

Prioritization (Simple Scoring)

Use a lightweight score to avoid backlog paralysis:

• Decision impact
• Knowledge gap
• Timing urgency
• Feasibility (recruitment + time)

Repository & Taxonomy

• Store each study with: method, date, product area, roles, tasks, key findings, raw evidence links.
• Tag for reuse: problem type (navigation/forms/performance), component/pattern, funnel step.
• Prefer “atomic” findings (one insight per card) to enable recombination [Inference].

Consent, PII, and Access Control

Follow applicable privacy laws; GDPR is a primary reference for EU processing https://eur-lex.europa.eu/eli/reg/2016/679/oj

PII handling checklist:

• [ ] Collect minimum PII needed for scheduling and incentives.
• [ ] Store identity/contact separately from study data.
• [ ] Redact names/emails from transcripts before broad sharing.
• [ ] Restrict raw recordings to need-to-know access.
• [ ] Document consent, purpose, retention, and opt-out path.

Research Democratization (2026 Trend)

Research democratization is a recurring 2026 trend: non-researchers increasingly conduct research. Enable carefully with guardrails.

| Approach | Guardrails | Risk Level | |----------|------------|------------| | Templated usability tests | Script + task templates provided | Low | | Customer interviews by PMs | Training + review required | Medium | | Survey design by anyone | Central review + standard questions | Medium | | Unsupervised research | Not recommended | High |

Guardrails for non-researchers:

• [ ] Pre-approved research templates only
• [ ] Central review of findings before action
• [ ] No direct participant recruitment without ops approval
• [ ] Mandatory bias awareness training
• [ ] Clear escalation path for unexpected findings

Researching Non-Technical User Segments (2026)

Quick checklist for research involving users with low digital literacy or low tech confidence. Full guidance in references/non-technical-user-research.md.

• [ ] Assess digital literacy tier (excluded → dependent → hesitant → capable → confident)
• [ ] Recruit via offline-first channels (community centers, libraries, phone outreach)
• [ ] Use plain-language screening questions (no jargon, no self-rating scales)
• [ ] Adapt methods: moderated-only testing, shorter sessions (30-40 min), read tasks aloud
• [ ] Measure: unassisted task completion (>=80%), time-to-first-value (<2 min), error recovery rate
• [ ] Frame findings as "inclusion improvements," not "dumbing down"
• [ ] Cross-reference with simplification audit template

Measurement & Decision Quality (Core)

Research ROI Quick Reference

| Research Activity | Proxy Metric | Calculation | |-------------------|--------------|-------------| | Usability testing finding | Prevented dev rework | Hours saved × $150/hr | | Discovery interview | Prevented build-wrong-thing | Sprint cost × risk reduction % | | A/B test conclusive result | Improved conversion | (ΔConversion × Traffic × LTV) - Test cost | | Heuristic evaluation | Early defect detection | Defects found × Cost-to-fix-later |

Rules of thumb:

• 1 usability finding that prevents 40 hours of rework = $6,000 value
• 1 discovery insight that prevents 1 wasted sprint = $50,000-100,000 value
• Research that improves conversion 0.5% on 100k visitors × $50 LTV = $25,000/month

When NOT to Run A/B Tests

| Situation | Why it fails | Better method | |----------|--------------|---------------| | Low power/traffic | Inconclusive results | Usability tests + trends | | Many variables change | Attribution impossible | Prototype tests → staged rollout | | Need “why” | Experiments don’t explain | Interviews + observation | | Ethical constraints | Harmful denial | Phased rollout + holdouts | | Long-term effects | Short tests miss delayed impact | Longitudinal + retention analysis |

Common Confounds (Call Out Early)

Always check for these in experiment evaluations. List each relevant confound with its risk level and how to verify — do not just name them:

• Selection bias (only power users respond) — check segment composition.
• Survivorship bias (you miss churned users) — compare with cohort-level data.
• Novelty effect (short-term lift) — plot daily metrics to check for trend decay.
• Instrumentation changes mid-test (metrics drift) — confirm no concurrent deployments.
• Sample ratio mismatch (SRM) — run chi-square on assignment counts.
• Peeking / multiple looks — confirm test was not checked before pre-set end date.
• Feature interaction — check if other experiments ran concurrently on same surface.

Optional: AI/Automation Research Considerations

Use only when researching automation/AI-powered features. Skip for traditional software UX.

2026 benchmark: Trend reports consistently highlight AI-assisted analysis. Use AI for speed while keeping humans responsible for strategy and interpretation. Example reference: https://www.lyssna.com/blog/ux-research-trends/

Key Questions

| Dimension | Question | Methods | |----------|----------|---------| | Mental model | What do users think the system can/can’t do? | Interviews, concept tests | | Trust calibration | When do users over/under-rely? | Scenario tests, log review | | Explanation usefulness | Does “why” help decisions? | A/B explanation variants, interviews | | Failure recovery | Do users recover and finish tasks? | Failure-path usability tests |

Error Taxonomy (User-Visible)

| Failure type | Typical impact | What to measure | |-------------|----------------|----------------| | Wrong output | Rework, lost trust | Verification + override rate | | Missing output | Manual fallback | Fallback completion rate | | Unclear output | Confusion | Clarification requests | | Non-recoverable failure | Blocked flow | Time-to-recovery, support contact |

Optional: AI-Assisted Research Ops (Guardrailed)

• Use automation for transcription/tagging only after PII redaction.
• Maintain an audit trail: every theme links back to raw quotes/clips.

Synthetic Users: When Appropriate (2026)

Trend reports frequently mention synthetic/AI participants. Use with clear boundaries. Example reference: https://www.lyssna.com/blog/ux-research-trends/

| Use Case | Appropriate? | Why | |----------|--------------|-----| | Early concept brainstorming | WARNING: Supplement only | Generate edge cases, not validation | | Scenario/edge case expansion | PASS Yes | Broaden coverage before real testing | | Moderator training/practice | PASS Yes | Practice without participant burden | | Hypothesis generation | PASS Yes | Explore directions to test with real users | | Validation/go-no-go decisions | FAIL Never | Cannot substitute lived experience | | Usability findings as evidence | FAIL Never | Real behavior required | | Quotes in reports | FAIL Never | Fabricated quotes damage credibility |

Critical rule: Synthetic outputs are hypotheses, not evidence. Always validate with real users before shipping.

Navigation

Resources

Core Research Methods:

• references/research-frameworks.md — JTBD, Kano, Double Diamond, Service Blueprint, opportunity mapping
• references/ux-audit-framework.md — Heuristic evaluation, cognitive walkthrough, severity rating
• references/usability-testing-guide.md — Task design, facilitation, analysis
• references/ux-metrics-framework.md — Task metrics, SUS/HEART, measurement guidance
• references/customer-journey-mapping.md — Journey mapping and service blueprints
• references/pain-point-extraction.md — Feedback-to-themes method
• references/review-mining-playbook.md — B2B/B2C review mining

Demographic & Quantitative Research:

• references/demographic-research-methods.md — Inclusive research for seniors, children, cultures, disabilities
• references/non-technical-user-research.md — Research methods for non-technical and low-digital-literacy users
• references/ab-testing-implementation.md — A/B testing deep-dive (sample size, analysis, pitfalls)

Competitive UX Analysis & Flow Patterns:

• references/competitive-ux-analysis.md — Step-by-step flow patterns from industry leaders (Wise, Revolut, Shopify, Notion, Linear, Stripe) + benchmarking methodology

Research Operations & Methods:

• references/research-repository-management.md — Repository architecture, taxonomy, atomic research, PII handling, adoption metrics
• references/survey-design-guide.md — Question types, bias prevention, sampling, sample size, distribution, platform comparison
• references/remote-research-patterns.md — Moderated remote, unmoderated testing, async methods, recruitment, tool comparison

Feedback Collection & Analysis:

• references/bigtech-feedback-patterns.md — How top companies collect and act on user feedback
• references/feedback-tools-guide.md — Feedback collection tool setup guides and selection matrix

Evaluative Iteration:

• references/evaluative-research-loop.md — Prototype-parity polishing loop (two-surface audit, drift classification, fast iteration)

Data & Sources:

• data/sources.json — Curated external references

Domain-Specific UX Benchmarking

IMPORTANT: When designing UX flows for a specific domain, you MUST use WebSearch to find and suggest best-practice patterns from industry leaders.

Trigger Conditions

• "We're designing [flow type] for [domain]"
• "What's the best UX for [feature] in [industry]?"
• "How do [Company A, Company B] handle [flow]?"
• "Benchmark our [feature] against competitors"
• Any UX design task with identifiable domain context

Domain → Leader Lookup Table

| Domain | Industry Leaders to Check | Key Flows | |--------|---------------------------|-----------| | Fintech/Banking | Wise, Revolut, Monzo, N26, Chime, Mercury | Onboarding/KYC, money transfer, card management, spend analytics | | E-commerce | Shopify, Amazon, Stripe Checkout | Checkout, cart, product pages, returns | | SaaS/B2B | Linear, Notion, Figma, Slack, Airtable | Onboarding, settings, collaboration, permissions | | Developer Tools | Stripe, Vercel, GitHub, Supabase | Docs, API explorer, dashboard, CLI | | Consumer Apps | Spotify, Airbnb, Uber, Instagram | Discovery, booking, feed, social | | Healthcare | Oscar, One Medical, Calm, Headspace | Appointment booking, records, compliance flows | | EdTech | Duolingo, Coursera, Khan Academy | Onboarding, progress, gamification |

Required Searches

When user specifies a domain, execute:

1. Search: "[domain] UX best practices 2026"
2. Search: "[leader company] [flow type] UX"
3. Search: "[leader company] app review UX" site:mobbin.com OR site:pageflows.com
4. Search: "[domain] onboarding flow examples"

What to Report

After searching, provide:

• Pattern examples: Screenshots/flows from 2-3 industry leaders
• Key patterns identified: What they do well (with specifics)
• Applicable to your flow: How to adapt patterns
• Differentiation opportunity: Where you could improve on leaders

Example Output Format

DOMAIN: Fintech (Money Transfer)
BENCHMARKED: Wise, Revolut

WISE PATTERNS:
- Upfront fee transparency (shows exact fee before recipient input)
- Mid-transfer rate lock (shows countdown timer)
- Delivery time estimate per payment method
- Recipient validation (bank account check before send)

REVOLUT PATTERNS:
- Instant send to Revolut users (P2P first)
- Currency conversion preview with rate comparison
- Scheduled/recurring transfers prominent

APPLY TO YOUR FLOW:
1. Add fee transparency at step 1 (not step 3)
2. Show delivery estimate per payment rail
3. Consider rate lock feature for FX transfers

DIFFERENTIATION OPPORTUNITY:
- Neither shows historical rate chart—add "is now a good time?" context

Trend Awareness Protocol

IMPORTANT: When users ask recommendation questions about UX research, you MUST use WebSearch to check current trends before answering.

Tool/Trend Triggers

• "What's the best UX research tool for [use case]?"
• "What should I use for [usability testing/surveys/analytics]?"
• "What's the latest in UX research?"
• "Current best practices for [user interviews/A/B testing/accessibility]?"
• "Is [research method] still relevant in 2026?"
• "What research tools should I use?"
• "Best approach for [remote research/unmoderated testing]?"

Tool/Trend Searches

1. Search: "UX research trends 2026"
2. Search: "UX research tools best practices 2026"
3. Search: "[Maze/Hotjar/UserTesting] comparison 2026"
4. Search: "AI in UX research 2026"

Tool/Trend Report Format

After searching, provide:

• Current landscape: What research methods/tools are popular NOW
• Emerging trends: New techniques or tools gaining traction
• Deprecated/declining: Methods that are losing effectiveness
• Recommendation: Based on fresh data and current practices

Example Topics (verify with fresh search)

• AI-powered research tools (Maze AI, Looppanel)
• Unmoderated testing platforms evolution
• Voice of Customer (VoC) platforms
• Analytics and behavioral tools (Hotjar, FullStory)
• Accessibility testing tools and standards
• Research repository and insight management

Templates

• Shared plan template: ../software-clean-code-standard/assets/checklists/ux-research-plan-template.md — Product-agnostic research plan template (core + optional AI)
• assets/research-plan-template.md — UX research plan template
• assets/testing/usability-test-plan.md — Usability test plan
• assets/testing/usability-testing-checklist.md — Usability testing checklist
• assets/audits/heuristic-evaluation-template.md — Heuristic evaluation
• assets/audits/ux-audit-report-template.md — Audit report

Evaluative Research Loop

For prototype-parity polishing (fast iteration when product is "almost ideal"), see references/evaluative-research-loop.md. Covers: two-surface audit, drift classification (layout/density/control/content/state), friction-based prioritization, banner/loading guardrails, localization-readiness checks, and fast iteration cadence.

Fact-Checking

• Use web search/web fetch to verify current external facts, versions, pricing, deadlines, regulations, or platform behavior before final answers.
• Prefer primary sources; report source links and dates for volatile information.
• If web access is unavailable, state the limitation and mark guidance as unverified.