UX Research with Claude: 32 Claude Skills

What these skills are

One user study, from the research plan to the repository entry, as installable Claude skills.

This is Claude for the work around the sessions: writing the plan that says which decision the study informs, getting real people booked and consented, writing guides that do not lead, capturing what happened the same day, coding nine transcripts into themes with every quote kept, and ending in a readout that asks a named person for a decision.

The skills follow one study in order, in seven stages, and each stage hands a named artifact to the next. Planning the study hands over what is already known with its dates, the assumption map that shows which bet carries the risk, the research plan, and an answer to the colleague offering synthetic users instead of real ones. Recruiting and consenting hands over the recruitment brief, the screener with its answer key, the consent form, the data handling plan, the incentive plan and the rules for the participant database. The guides hand over the interview guide, the moderated and unmoderated test plans, the benchmark, the survey and the diary study. Running and capturing hands over the contextual inquiry plan, sessions disabled people can take part in, the AI feature study and same-day debrief notes. Synthesis hands over the codebook and themes, the usability findings by severity, the survey analysis with honest base sizes, the insight statements and the evidence trace audit. Maps and opportunities hand over the journey map, the jobs to be done and the opportunity solution tree. Sharing and reuse hands over the readout, the action tracker, the repository entries and the tagging taxonomy.

The red line: Claude plans, organises and synthesises what real people said and did, with every insight traced to a quote or an observed session. It never invents a participant, a quote, a finding or a metric, and it never stands in for a user.

Quotes, counts and metrics come only from the sessions and responses you collected. Anything missing stays a bracketed placeholder or an open question rather than a plausible number, counts are written as “[n] of [N] participants”, single voices are listed apart and counter-examples stay in. Each skill uses the real mechanics of its method — research questions kept apart from interview questions, the screener's hidden qualifying answers and decoys, think-aloud and neutral prompts, the severity scale, the codebook passes, the importance and evidence axes — rather than a generic gather, analyse, recommend.

Problems are rated; people are not. Severity rates usability problems, the screener checks fit to the study brief and a person decides, and no skill scores, ranks or profiles a participant. Anonymise transcripts and recordings before you paste them. Consent, privacy and data retention points come back as questions ending in “check with your privacy lead or a qualified adviser”, never as advice. And the synthetic user check exists so that an AI persona's answers are matched against real evidence instead of quietly becoming a finding.

Each skill is one named method or one artifact: the inputs it needs, where it breaks, a fixed output template, a finish line and the next skill to run. Where a skill fits a Claude surface it says so — Claude Docs for plans, consent forms and memos, Claude Slides for the readout, Claude Design for the journey map visual, a Project to hold the study's files across weeks, and connectors such as Figma, a drive or your team's tracker to read what already exists. Connectors read; you send every message yourself.

Mechanically, each skill is one folder with a SKILL.md file. The 32 are grouped into the seven stages, and each one gives full value on its own: run one, run a stage, or work one study through from the plan to the repository entry.

Download all 32 skills. One zip: ready-to-install skill zips, readable SKILL.md files, and the sources behind every method. Free, no signup. Download the zip · Browse on GitHub

The 32 skills you get

The set follows one study in seven stages: planning it, recruiting and consenting real people, writing the interview and test guides, running and capturing the sessions, synthesising with every quote kept — then the maps and opportunities, and sharing findings someone can still find next year.

1 · Plan the study

1. Desk Research Summary

Use when: A study is about to start and nobody has read last year's research

Output: What is already known with source and date, what is stale, open questions, what not to research again

2. Assumption Map

Use when: Research gets cut and you need to show which bet the team makes without evidence

Output: Assumptions placed by importance and evidence, risky unknowns to research first, what a skipped study puts at risk

3. UX Research Plan

Use when: You have two weeks and a stakeholder question, and need a plan signed before recruiting

Output: Decision informed, research questions, method and why, sample, timeline, roles, risks, what it will not answer

4. Synthetic User Check

Use when: Someone says “just ask the AI personas” or pastes synthetic answers into the deck

Output: Each synthetic claim matched to real evidence, what stays a hypothesis, what is rejected, the real study that would answer it

2 · Recruit and consent

5. Participant Recruitment Brief

Use when: Sessions are next week and nobody owns getting real people booked

Output: Who to hear from in behaviour terms, who wastes a slot, quotas, channels, owner, booking tracker, no-show buffer, fraud checks

6. Participant Screener

Use when: The screener reads like a quiz, or last round's participants were not who they said

Output: Behaviour and recency questions, hidden qualifying answers, decoys, disqualifiers, quotas, answer key for fit to the brief

7. Informed Consent Form

Use when: An AI notetaker will join the call and the consent form predates it

Output: Plain-language information sheet, consent items, remote and verbal script, accessible version

8. Research Data Handling Plan

Use when: Recordings sit in personal drives and nobody knows when they get deleted

Output: Data inventory, minimum to collect, storage, access, anonymise steps, special category flags, deletion dates

9. Participant Incentive Plan

Use when: You promised a thank-you and finance asks how, when and whether it is taxable

Output: Incentive per study type with amounts left blank, payment method and timing, no-show rules, questions for finance

10. Participant Database Plan

Use when: The team keeps recruiting the same five friendly customers from a spreadsheet nobody owns

Output: Minimum fields, consent to recontact, contact limits and rest periods, opt-out and deletion, access, contact log

3 · Interview and test guides

11. User Interview Guide

Use when: Interviews are booked and the question list starts with “would you use”

Output: Research questions mapped to interview questions, story openers, probes, time budget, leading-question check

12. Usability Test Plan

Use when: The prototype is ready and the script asks “do you like it”

Output: Test goals, task scenarios with success set in advance, think-aloud, moderator script, observer grid, pilot

13. Unmoderated Test Plan

Use when: You need ten sessions by Friday and nobody can moderate

Output: Self-explaining tasks, success per task, in-tool screener, attention checks, device pilot, what not to test this way

14. Usability Benchmark Study

Use when: Leadership asks whether the redesign made things better and you have no baseline

Output: Same tasks each round, task success, time, SEQ and SUS defined before data, baseline, re-run date

15. UX Survey Design

Use when: A “quick survey” is about to go out with questions that start “don't you agree”

Output: Decision served, one construct per question, neutral wording, scales, order, pilot, what it can and cannot say

16. Diary Study Plan

Use when: The behaviour happens across weeks and nobody can recall it in an interview

Output: Behaviour over time, entry prompts, cadence, length, reminders, drop-off plan, onboarding call, coding plan

4 · Run and capture

17. Contextual Inquiry Plan

Use when: What users say and what they do look like two different stories

Output: Where and when to observe, apprentice stance, what to watch, note grid, site permissions, debrief per visit

18. Accessible Research Session Plan

Use when: No disabled person has ever been in a session

Output: Recruiting disabled and older participants, access needs in advance, their own assistive technology, accessible consent, adjustments

19. AI Feature User Study

Use when: An AI feature ships next quarter and nobody has watched a user meet a wrong answer

Output: Current mental model, expectations, tasks on real inputs, wrong-answer scenarios, trust and correction checks, Wizard of Oz option

20. Interview Debrief Notes

Use when: You just finished three calls in a row and the notes are a wall of text

Output: Same-day notes per session, verbatim quotes with timestamps, observation vs interpretation, surprises, next questions

5 · Synthesise with quotes kept

21. Thematic Analysis

Use when: Nine interviews are done and the team wants themes by tomorrow

Output: Codebook, codes with quotes and participant ids, themes with counts, counter-examples, single-voice observations

22. Usability Test Findings

Use when: The tests “went well” and the notes say four people failed the core task

Output: Did vs said per task, completion with and without help, issues by severity, rainbow sheet, clips to pull

23. Survey Results Analysis

Use when: The export is in and someone is about to average a Likert scale into a headline

Output: Who answered, cleaning log, frequencies with base sizes, cross-tabs where the base allows, open-text coding, limits

24. Research Insight Statements

Use when: You have themes and the room still says “so what”

Output: Five to nine “want X but do Y because Z” insights with quotes, counts and confidence, confirmations apart

25. Evidence Trace Audit

Use when: The deck goes to leadership tomorrow and nobody checked where each sentence came from

Output: Every claim traced to a quote, session or data row, flags for untraced, thin, misquoted or synthetic claims

6 · Maps and opportunities

26. User Journey Map

Use when: Every team owns one touchpoint and nobody sees where the experience breaks

Output: Stages, actions, thoughts in participants' words, feelings, pain points, each cell tagged to evidence, gaps marked

27. Jobs to Be Done Statements

Use when: The team describes features and cannot say what progress people are trying to make

Output: Jobs with situation, motivation and outcome, job stories, switching forces, each tied to quotes

28. Opportunity Solution Tree

Use when: Research produced twenty needs and the team jumps to the first idea

Output: Outcome, opportunities from interviews, ideas per opportunity, assumption test for each, branch to explore next

7 · Share and reuse

29. Research Readout Deck

Use when: Findings get a nice meeting and then nothing changes

Output: Decision first, three to five insights with quote and count, what we did not learn, recommendations, the decision asked

30. Research Action Tracker

Use when: Six months later nobody can say what the last study changed

Output: Each recommendation with owner, decision and reason, date, evidence, follow-up check, what changed

31. Research Repository Entry

Use when: The study is done and its findings will die in a slide deck

Output: Atomic entries (experiment, fact, insight, recommendation), conditions and date, anonymised quote, tags, review date

32. Research Tagging Taxonomy

Use when: The repository has hundreds of tags and nobody finds anything

Output: A short set of broad tags with definitions, rules for new tags, owner, a label test, review cadence

Where to start, and how the skills chain

Start with the research plan. Almost everything downstream reads it: the recruitment brief turns its sample into quotas, the screener checks fit against it, the interview guide maps its research questions onto questions you can actually ask, and the readout answers the decision it named. After that there is no compulsory sequence — bring the thing that is in front of you. Three calls this morning? Start at Interview Debrief Notes. Nine transcripts waiting? Start at Thematic Analysis. Readout on Thursday? Start at the Research Readout Deck, which will send you back for insight statements if you do not have them. Every skill lists the inputs it needs, so it tells you when the step before it is the one that is actually missing.

They chain. uxr-desk-research says what is already known and uxr-assumption-map says which belief carries the risk, which is what uxr-research-plan needs to pick a method and a sample; uxr-synthetic-user-check is for the colleague offering AI personas instead. uxr-recruitment-brief, uxr-participant-screener, uxr-consent-form, uxr-data-handling-plan, uxr-incentive-plan and uxr-participant-database get real people booked, consented and looked after. uxr-interview-guide, uxr-usability-test-plan, uxr-unmoderated-test-plan, uxr-usability-benchmark, uxr-survey-design and uxr-diary-study write the instrument, and uxr-contextual-inquiry, uxr-accessible-sessions, uxr-ai-feature-study and uxr-interview-debrief run and capture the sessions. uxr-thematic-analysis, uxr-usability-findings and uxr-survey-analysis turn that into themes, severity and frequencies, uxr-insight-statements answers “so what”, and uxr-evidence-trace-audit checks every claim before the deck leaves the room. uxr-journey-map, uxr-jtbd-statements and uxr-opportunity-tree turn findings into where the experience breaks and what to work on next. Then uxr-readout-deck, uxr-action-tracker, uxr-repository-entry and uxr-tagging-taxonomy make the study end in a decision and stay findable.

The sources behind each method are in resources/evidence-and-sources.md, inside the pack.

Setup guide

1. Download the pack. One zip: an install folder with 32 ready-to-upload skill zips, a skills folder with the same 32 skills as readable SKILL.md files, and resources/evidence-and-sources.md, which names the source behind each method.
2. Install your skills. In Claude Code, add the marketplace and install ux-research-pack, and all 32 load at once. In Claude, turn on code execution in Settings, then Capabilities, then go to Customize, then Skills and upload one zip per skill from the install folder. Prefer working from files? Add the SKILL.md files to your Project knowledge instead; it works, just less cleanly. Anonymise transcripts and recordings before you paste them, and keep participant data out of Claude memory.
3. Start with the study in front of you. Starting fresh? Run the UX Research Plan with the stakeholder question and your deadline, so every later skill knows which decision the study informs. Already mid-study? Bring what is actually happening: the three calls you finished this morning, the nine transcripts waiting to be coded, the readout on Thursday. Each skill lists the inputs it needs and names the next skill to run, so one answer turns into the chain on its own.

Where to start

The quality bar

Every skill in the pack holds the same standard:

• One method, applied properly: the research questions kept apart from interview questions, the screener decoys and answer key, the think-aloud prompts, the severity scale, the codebook passes, the importance and evidence axes — not a generic “gather, analyse, recommend”
• Every insight traced: each theme, insight and map cell names the quote, the session or the data row behind it, with counts written as “[n] of [N] participants”. Single voices are listed apart, and counter-examples stay in
• No invented numbers: no made-up participants, quotes, scores, sample sizes, incentive amounts or benchmarks. A metric appears only when it came from sessions or responses you collected; anything missing becomes a bracketed placeholder or an open question
• Problems are rated, people are not: severity rates usability problems, the screener checks fit to the study brief and a person decides, and no skill scores, ranks or profiles a participant
• Participant data handled with care: anonymise transcripts and recordings before pasting, and consent, privacy and data points end with “check with your privacy lead or a qualified adviser” — the skills ask the questions and never answer them as advice
• The red line: Claude plans, organises and synthesises what real people said and did, with every insight traced to a quote or an observed session; it never invents a participant, a quote, a finding or a metric, and it never stands in for a user

Who made this

Polar Bear is a people ops consultancy for human-size teams (20 to 200 people). Built by ex-McKinsey founders with a dream to make AI work for People, not instead of them. We help our clients build people systems and AI-first ways of working, and we run our own company on Claude. This pack is the free, self-serve version of how we work.

This pack is for whoever is accountable for the study. If you run a design or research practice and want it working this way — AI carrying the plans, the coding and the readouts so people do the part only people can do — that's what we build with clients.

Meet Pauline. Want your research practice working this way? Bring the stage you are stuck on and we will pick the skills for it. Book a 30-minute call

Install in Claude Code

Two lines, and every skill in the pack loads at once.

/plugin marketplace add polar-bear-org/claude-skills
/plugin install ux-research-pack@polar-bear-skills

Using Claude on the web instead? Download the zip and upload each skill from its install folder under Customize → Skills.