ux-research — команда для Claude Code: что делает и как установить
Команда ux-research планирует пользовательские исследования, генерирует сценарии интервью и синтезирует собранные заметки в продуктовые выводы.
MIT ★ 110 обновлён 2026-11-02
/plugin marketplace add Laith0003/ux-skill
/plugin install ux@ux-skillЧто делает ux-research
Команда /ux-research входит в плагин ux и помогает подготовить исследование пользователей без методологических перекосов. В режиме планирования она классифицирует вопрос по типу задачи (поиск проблемы, валидация решения, юзабилити-тест), подбирает размер выборки и генерирует артефакты: скрипты глубинных интервью, опросы со случайным порядком вариантов или скринеры для рекрутинга.
Команда требует опираться на прошлый опыт респондентов. Формулировки вроде «пользовались бы вы этим» отсекаются правилами генерации. После завершения сессий флаг --synthesize запускает обработку заметок или CSV-файлов через задачу research-synthesizer. Результаты исследования и параметры выборки фиксируются в файле .ux/last-research.json.
Какие задачи решает
- составить нейтральный сценарий интервью без наводящих вопросов
- спроектировать опрос с проверками внимания и рандомизацией ответов
- подготовить анкету скрининга и рассчитать бюджет на рекрутинг
- разобрать транскрипты и заметки через флаг
--synthesize
Примеры запросов агенту
- /ux-research составь план юзабилити-теста для формы оплаты, бюджет 10 сессий по 30 минут
- /ux-research подготовь скрипт интервью для поиска проблем в онбординге
- /ux-research --synthesize путь к заметкам notes.txt, вопрос: почему отваливаются пользователи
Как установить ux-research
1Через маркетплейс плагинов
/plugin marketplace add Laith0003/ux-skill/plugin install ux@ux-skill2Вручную: положить файлы
.claude/commands/ux-research.mdВызов — /ux-research. Для всех проектов — ~/.claude/commands/
Вручную: положить файлы
~/.codex/prompts/ux-research.mdВызов — /prompts:ux-research. В Codex этот формат помечен устаревшим, замена — скиллы
Вручную: положить файлы
.github/prompts/ux-research.prompt.mdВручную: положить файлы
.gemini/commands/ux-research.tomlGemini хранит команды в TOML: текст запроса — в поле prompt
Вопросы об ux-research
Как установить ux-research в Claude Code?
Подключите маркетплейс командой /plugin marketplace add Laith0003/ux-skill, а затем установите плагин вызовом /plugin install ux@ux-skill.
Что делает команда ux-research?
Она определяет метод исследования, формирует рабочий артефакт (скрипт, опросник, скринер) и задаёт план рекрутинга с анализом данных.
Как переключить команду в режим обработки результатов?
Запустите /ux-research --synthesize, передав исходный исследовательский вопрос и путь к файлам с сырыми заметками или CSV-выгрузкой опроса.
Исходный текст
ux-research.md · MIT
---
description: User research planning — interview scripts, survey design, recruitment screener, synthesis. Triggers on "plan a research study", "I need interview questions", "design a survey", "how do I recruit users". Use when planning user research, writing interview scripts, designing a survey, creating a screener, scoping recruitment, synthesizing research data into recommendations. Skip when the user already has research data and wants synthesis (use ux-research --synthesize), the answer is already known with high confidence, low-risk reversible decisions, backend or infrastructure work.
allowed-tools: Read, Write, Edit, Bash(ls:*), Bash(cat:*), Bash(find:*), Bash(mkdir:*), Glob, Grep, Task
disable-model-invocation: false
---
# /ux-research
You are running the `/ux-research` command from the `ux` plugin. The job is to plan a research study and produce the artifacts to run it — or, with `--synthesize`, digest results into actionable findings. Methodologically rigorous, anti-bias by default.
## When to use
Triggers: "plan a research study", "I need interview questions", "design a survey", "how do I recruit users", "user testing plan", "diary study", "preference test", "fake door", "smoke test", "synthesize my interview notes". Also runs as a sub-step of `/ux-workshop` between phases.
## Input
Required (planning mode):
- **Research question(s)**: what you want to learn, in plain language
- **Design decision the research informs**: the action that will change based on findings
- **Budget / timeline**: minutes per session, total sessions, dollars
Required (synthesis mode, triggered by `--synthesize`):
- Path to raw notes / transcripts / survey CSV
- The original research question(s) for grounding
If anything's missing, ask once: *"Research question, the design decision it informs, and budget/timeline (sessions × minutes × incentive)?"*
## Process (planning mode)
### 1. Frame the research question
Classify the question into one of:
- **Problem discovery** — is there a problem worth solving here?
- **Solution validation** — does our proposed solution address the problem?
- **Usability testing** — can users actually use what we built?
- **Preference / desirability** — which option resonates?
- **Segmentation** — who are the user types and what differentiates them?
A research question that fits two categories is too broad. Split it.
### 2. Pick the method
| Question type | Default method | Alternative |
|---|---|---|
| Problem discovery | Semi-structured interview (n=5-8) | Diary study (n=5, 1 week) |
| Solution validation | Concept test (n=5-8) + structured interview | Fake door (n=∞ via traffic) |
| Usability testing | Moderated task-based test (n=5-8) | Unmoderated remote (n=15+) |
| Preference / desirability | Within-subjects preference test (n=8-12) | A/B test (n=large, traffic-dependent) |
| Segmentation | Survey (n=100+) | Card sort + cluster analysis (n=15-25) |
Smoke tests and fake doors are valid for binary problem-existence checks before deeper qual work.
### 3. Generate the artifact
Build the deliverable for the chosen method:
- **Interview script** — open-ended + behavior-focused questions, no leading wording. Structure: warm-up (2-3 q), context (current behavior), specific (past experience with the problem), reflection (what would solve this), wrap-up. Never ask "would you use" — ask about past behavior.
- **Survey** — anti-bias structure. Randomize answer order. Mix 5-point Likert with open-ended. Screen out professionals/insiders. Attention check question.
- **Usability test plan** — tasks framed as goals not steps. Time-box per task. Define success criteria upfront. Think-aloud protocol stated to participant.
- **Diary study brief** — daily prompt, max 5 min effort, end-of-study debrief interview.
- **Screener questionnaire** — disqualifying criteria first, then qualifying. Stated incentive, time commitment, schedule.
### 4. Recruitment plan
- **Where**: user pool / customer list / Respondent.io / UserInterviews / Prolific / community forums / DM intercept on the product / partner contacts
- **Incentive**: dollar value matched to time (industry rule of thumb: ~$1-2/minute for B2C, $3-6/minute for B2B / experts)
- **Volume**: target sample + 30% buffer for no-shows
### 5. Analysis plan
State upfront how findings will be synthesized:
- Affinity mapping for interviews
- Cross-tabs + chi-square for surveys
- Task completion + time-on-task + System Usability Scale (SUS) for usability
- Tag taxonomy for diary studies
## Process (synthesis mode)
Dispatch `research-synthesizer` via Task. Pass:
- Raw input files
- Original research question
- Method that was used
- Instruction to produce: themes (with supporting quotes), behavioral patterns, surprises, decision-ready insights (each tied back to the original design decision), recommended next steps
### Output format
```
─── research plan ───
Question: <verbatim>
Decision: <what changes based on findings>
Method: <method>
Sample: n=<target> (n=<target+buffer> recruited)
Timeline: <weeks>
Budget: $<total>
─── artifact ───
<the deliverable — interview script / survey / test plan / screener — in full>
─── recruitment ───
Source: <where>
Incentive: $<amount> / <duration>
Screener: <key criteria>
─── analysis plan ───
<how findings will be synthesized>
─── next ───
Run the study, then come back with /ux-research --synthesize
```
### State
Write to `.ux/last-research.json`:
```json
{
"command": "ux-research",
"timestamp": "<ISO8601>",
"mode": "planning|synthesis",
"question": "<verbatim>",
"decision": "<design decision>",
"method": "<method>",
"sample_target": <n>,
"incentive": "<amount>",
"artifact_file": "<path>",
"synthesis_findings": ["<key themes>"]
}
```
## Hard rules
- Never ask "would you use" — ask about past behavior. Stated intent is a noise channel.
- Never recommend n<5 for qual; never recommend n<30 for quant; never recommend n<100 for segmentation surveys.
- Screen out product professionals (designers, PMs, UXers) unless they ARE the target user.
- Randomize answer order in surveys. No order-bias artifacts.
- Attention check questions on every survey.
- Time-box every session. Sessions that run long produce diminishing returns and respondent fatigue.
- Tie every finding back to the original research question. Findings that don't map are notes, not insights.
## Failure modes
- **Multi-question studies**: trying to learn 5 things in one session = learning 0 things well. Split the study.
- **Leading wording**: "How much do you love X?" — reject and rewrite.
- **Recruiting from product users only**: confirmation bias. Mix in non-users for discovery work.
- **No incentive**: response rate craters. Always pay participants.
- **Synthesis as a quote dump**: themes need to be patterns across participants, not "one person said X."
## Error Handling
| Error condition | Recovery |
|---|---|
| No research budget specified | Ask: "Sessions × minutes × incentive — what's the budget?" |
| Conflicting research goals (study tries to learn 3+ things) | List the goals, ask the user to pick the primary one; offer to split into separate studies |
| Recruitment channel unclear | Propose channels matched to the audience (Respondent.io / UserInterviews / Prolific / customer list / community / DM intercept) |
| Synthesis mode: raw input files missing or unreadable | Ask for the file path or pasted transcript; stop until provided |
| Sample size requested below floor (n<5 qual, n<30 quant) | Refuse and explain the floor; offer the minimum viable design |
| Voice/leading wording in interview script | Rewrite and flag the bias risk before delivery |
For path issues: see references/process/discovery-protocol.md for state file location (.ux/ in project root). Report bugs at https://github.com/Laith0003/ux-skill/issues.
## Next prompt
After `/ux-research` (planning):
- Run the study
- `/ux-research --synthesize` — dispatch `research-synthesizer` on raw results
After `/ux-research --synthesize`:
- `/ux-design` — apply findings to a design
- `/ux-discover --frame`: re-baseline the brief on what you learned
- `/ux-workshop` — if findings open a bigger question
- `/ux-next` — let the conductor pick
---
## v2 Python integration
Research inputs (interviews, analytics, A/B results, support tickets, competitive sites) get synthesized into a Brief object that feeds the recommender.
### Synthesize research into recommender-ready Brief
After collecting and reading research artefacts, the LLM's job is to extract:
- `audience` (specific, not generic)
- `tone` (3-5 words from what users said matters)
- `must_have` (from analytics + tickets — pain points fixed = required)
- `forbidden` (from competitive sites — patterns to avoid)
- `reference_brands` (from competitive analysis — what users praised)
Write the synthesized brief:
```bash
mkdir -p .ux
cat > .ux/last-discovery.json <<JSON
{
"answers": {
"project_type": "<extracted>",
"audience": "<from interviews/analytics>",
"primary_goal": "<top job-to-be-done>",
"tone": "<from what users praised>",
"must_have": "<analytics pain points + tickets>",
"forbidden": "<from competitor anti-patterns>",
"reference_brands": "<from competitive praise>",
"stack": "<inferred or asked>",
"region": "<from user data>",
"success_metric": "<from analytics baseline>"
}
}
JSON
```
### Verify the brief produces a sensible recommendation
```bash
python3 -m engine.cli.main --no-pretty recommend --brief-file=.ux/last-discovery.json | python3 -c "
import json, sys
r = json.load(sys.stdin)
for line in r.get('rationale', []):
print(line)
"
```
If the recommender's rationale doesn't match what research said, the brief synthesis was wrong — iterate.
### Fallback
If the engine isn't available, produce the brief as a markdown document with the 10 fields filled in and recommend `/ux-discover` next.
© Laith Aljunaidy · MIT · Laith0003/ux-skill