snqb/bishkek-report

Bishkek Report

You are about to assemble a focus group from Kyrgyzstan. They don't know each other, they didn't sign up for this, and they have opinions.

When to use

• You built something and want to know if real people would survive it
• You need UX feedback that isn't from a designer in San Francisco
• You want to stress-test your app against the full spectrum: a 72-year-old эне on a Galaxy A13, a 16-year-old preparing for ОРТ on a Xiaomi, a developer with DevTools open, and a journalist who needs a quote in 5 minutes

The Cast

Generate 8–12 people that reflect real Kyrgyzstan demographics. Not stereotypes — actual variance:

| Axis | Examples of variance | |------|---------------------| | Age | 16 → 72. Include at least one teenager and one pensioner | | Location | Bishkek, Osh, Naryn, Jalal-Abad oblast village, Moscow diaspora, Ankara researcher | | Language | KY-only, RU-only, KY+RU, KY+RU+EN, TR+EN | | Tech level | ⭐ (nephew helps) → ⭐⭐⭐⭐⭐ (opens DevTools first) | | Profession | Teacher, student, developer, journalist, pensioner, activist, construction worker, marketolog, researcher | | Device | Xiaomi Redmi, Samsung Galaxy A-series, iPhone, MacBook, Windows laptop |

Each person gets a name, age, city, profession, languages, tech level. Keep it plausible. No one is named "Test User 4".

Process

1. Launch the app locally

# Start the server in tmux (never block main thread)
tmux has-session -t pi 2>/dev/null || tmux new-session -d -s pi
tmux new-window -d -t pi -n app 'cd PROJECT_DIR && START_COMMAND'
sleep 2
tmux capture-pane -t pi:app -p -S -5

2. Take screenshots with ABP (browser.js)

Use browser.js from the browser-testing skill. Capture at minimum:

• Home page (desktop)
• Home page (mobile 390×844)
• A search interaction + results
• A detail/content page (full page)
• Dark mode (if exists)
• Any "explore" / secondary page

B=~/.pi/agent/skills/browser-testing/browser.js

# Launch browser
$B start --headless

# Desktop screenshots (default viewport 1280x900)
$B nav 'http://localhost:PORT'
$B screenshot /tmp/review-home.png

# Interact — search, click, navigate. Capture each state.
$B click 'input[type="search"]'
$B type 'search query'
$B key Enter
$B screenshot /tmp/review-search.png

# Navigate to a detail page
$B click 'a.detail-link'
$B screenshot /tmp/review-detail.png --full

# Mobile viewport
$B nav 'http://localhost:PORT' --device 'iPhone 15'
$B screenshot /tmp/review-mobile.png

# Dark mode (if exists)
$B eval 'document.documentElement.classList.add("dark")'
$B screenshot /tmp/review-dark.png

3. Read screenshots

Use the read tool on each .png — actually look at the app. This is your eyes. Note:

• Visual hierarchy — what grabs attention first?
• Information density — too sparse? too cluttered?
• Typography — readable? too small?
• Touch targets — would a thumb hit them?
• Color/contrast — especially dark mode
• Broken layouts, overflow, clipped text

4. Hit the API

# Poke around the actual data
curl -s 'http://localhost:PORT/api/...' | python3 -m json.tool

Understand what the app does, not just what it looks like. Try edge cases:

• Empty search
• Very long input
• Non-Latin characters
• Features mentioned in the UI but possibly broken

5. Write the report

Each person gets:

## N. Name, age — profession, city

> [!quote] «One memorable sentence in their voice»

### Сценарий
What they did — specific actions, specific pages, specific searches.

### Что понравилось
- Bullet points. Specific. Reference actual UI elements, data, interactions.

### Что не понравилось
- Bullet points. Specific. Not vague "could be better" — what broke, what confused, what's missing.

### Оценка: **X/10**

> One-line summary in their voice.

6. Summary section

After all testimonies:

• Average score with ASCII bar chart
• Problem frequency — which issues hit the most people (use mermaid if Obsidian)
• Strengths table — what everyone agrees is good
• Prioritized issues — P0 (broken), P1 (significant), P2 (nice to have)
• One-sentence verdict

Output

• Format: .md with Obsidian features (callouts, mermaid, YAML frontmatter, tables)
• Language: Russian. The people are from Kyrgyzstan, they think in Russian (or KY, but you write in RU because your Kyrgyz is embarrassing)
• Location: .git/reports/user-testing-YYYYMMDD.md
• Tone: Honest. The people are not mean, but they're not impressed easily. Бүбүсара эне doesn't care about your proto-Turkic reconstruction if she can't read the font.

Cleanup

rm /tmp/review-*.png
tmux kill-window -t pi:app 2>/dev/null

Rules

1. Always screenshot — you must look at the app, not guess from code
2. Always hit the API — test what's behind the UI
3. People are specific — they search specific things, notice specific bugs, have specific needs
4. Scores vary — if everyone gives 7/10, your cast is too homogeneous
5. The grandmother and the developer should disagree — that's the point
6. Russian language — for the report text. UI elements and Kyrgyz terms quoted as-is
7. Clean up — screenshots, tmux windows, temp scripts