# Video Recording

Record browser automation sessions as MP4 video files. Uses CDP `Page.startScreencast` to capture JPEG frames and pipes them to ffmpeg for encoding.

## Prerequisites

**ffmpeg** must be installed. Choose один из вариантов:

### Вариант 1: в проект (рекомендуется)

Скачать essentials build с https://www.gyan.dev/ffmpeg/builds/, распаковать в `tools/ffmpeg/` проекта:

```
tools/ffmpeg/
├── bin/
│   ├── ffmpeg.exe      ← этот файл ищет startRecording()
│   ├── ffplay.exe
│   └── ffprobe.exe
└── ...
```

Код автоматически найдет `tools/ffmpeg/bin/ffmpeg.exe` - ничего больше настраивать не нужно.

### Вариант 2: глобально (один раз на машину)

Скачать, распаковать в любой каталог (напр. `C:\tools\ffmpeg`), добавить `bin/` в системный PATH.
После этого ffmpeg доступен во всех проектах.

### Вариант 3: через .v8-project.json (общий путь)

Чтобы не копировать ffmpeg в каждый проект, указать путь в конфиге:

```json
{
  "ffmpegPath": "C:\\tools\\ffmpeg\\bin\\ffmpeg.exe"
}
```

Модель прочитает это поле и передаст в `startRecording({ ffmpegPath })`.

### Порядок поиска ffmpeg

1. `opts.ffmpegPath` - явный путь (из `.v8-project.json` или параметра)
2. `FFMPEG_PATH` - переменная окружения
3. `ffmpeg` - в системном PATH
4. `tools/ffmpeg/bin/ffmpeg.exe` - относительно корня проекта

## API

### `startRecording(outputPath, opts?)`

Start recording the browser viewport to an MP4 file.

| Parameter | Type | Default | Description |
|-----------|------|---------|-------------|
| `outputPath` | string | required | Output .mp4 file path |
| `opts.fps` | number | 25 | Target framerate |
| `opts.quality` | number | 80 | JPEG quality (1-100) |
| `opts.ffmpegPath` | string | auto | Explicit path to ffmpeg binary |
| `opts.speechRate` | number | 70 | Ms per character for smart TTS wait. Increase for slower TTS providers (e.g. 85 for ElevenLabs) |

- Output directory is created automatically if it doesn't exist
- Throws if already recording or browser not connected
- Recording auto-stops when `disconnect()` is called

### `stopRecording()` → `{ file, duration, size, captions }`

Stop recording and finalize the MP4 file. Saves `.captions.json` next to the video if captions were collected.

| Return field | Type | Description |
|-------------|------|-------------|
| `file` | string | Absolute path to the MP4 file |
| `duration` | number | Recording duration in seconds |
| `size` | number | File size in bytes |
| `captions` | number | Number of captions collected during recording |

### `isRecording()` → boolean

Check if recording is active.

### `showCaption(text, opts?)`

Display a text overlay on the page (visible in recording). Calling again updates the text.

| Parameter | Type | Default | Description |
|-----------|------|---------|-------------|
| `text` | string | required | Caption text |
| `opts.position` | `'top'` \| `'bottom'` | `'bottom'` | Vertical position |
| `opts.fontSize` | number | 24 | Font size in px |
| `opts.background` | string | `'rgba(0,0,0,0.7)'` | Background color |
| `opts.color` | string | `'#fff'` | Text color |
| `opts.speech` | string \| false | - | TTS narration text. Omit = use displayed text, string = custom narration, false = skip narration |
| `opts.voice` | string | - | Per-caption voice override (provider-specific voice name/ID). Used by `addNarration` instead of the global voice |

When `text` is empty but `speech` is a string, the caption is still recorded for TTS (no visible overlay). Useful for narration-only captions (e.g. podcast mode).

The overlay uses `pointer-events: none` - does not interfere with clicking.

**Smart TTS wait** (during recording): `showCaption` automatically pauses for the estimated TTS speech duration (default ~70ms per character, min 2s; configurable via `startRecording({ speechRate })`). The next `wait()` call accounts for this - if the explicit pause is shorter than the TTS wait already done, no extra delay is added. If longer, only the remaining difference is waited. This means script authors don't need to calculate TTS timing manually.

### `hideCaption()`

Remove the caption overlay.

### `showTitleSlide(text, opts?)`

Display a full-screen title slide overlay (gradient background, centered text). Useful for intro/outro frames in video recordings. Calling again updates the content.

| Parameter | Type | Default | Description |
|-----------|------|---------|-------------|
| `text` | string | required | Title text (`\n` → line break) |
| `opts.subtitle` | string | `''` | Smaller text below the title |
| `opts.background` | string | dark gradient | CSS background |
| `opts.color` | string | `'#fff'` | Text color |
| `opts.fontSize` | number | 36 | Title font size in px |
| `opts.speech` | string \| false | - | TTS narration text. String = custom text, `true` = use title text, omit/false = no narration |
| `opts.voice` | string | - | Per-caption voice override for `addNarration` |

The overlay covers the entire viewport with `z-index: 999999` and `pointer-events: none`.

### `hideTitleSlide()`

Remove the title slide overlay.

### `showImage(imagePath, opts?)`

Display a full-screen image overlay (e.g. presentation slide screenshot). Reads the file, base64-encodes it, and renders as `<img>` in a fixed overlay - captured by CDP screencast automatically.

| Parameter | Type | Default | Description |
|-----------|------|---------|-------------|
| `imagePath` | string | required | Path to image file (PNG, JPG, GIF, WebP, SVG) |
| `opts.style` | `'blur'` \| `'dark'` \| `'light'` \| `'full'` | `'blur'` | Display style preset |
| `opts.background` | string | - | Custom background (overrides preset) |
| `opts.shadow` | boolean | preset | Show drop shadow on image |
| `opts.speech` | string \| false | - | TTS narration text while image is shown |
| `opts.voice` | string | - | Per-caption voice override for `addNarration` |

**Style presets:**
- `blur` - blurred+dimmed copy of the image as background, centered image with shadow
- `dark` - dark background (#2a2a2a) with shadow
- `light` - white background with shadow
- `full` - image fills entire screen (contain, no crop), black background, no shadow

Images are auto-scaled: small images scale up (min 50% of viewport), large images scale down (max 92%).

### `hideImage()`

Remove the image overlay.

### `setHighlight(on)`

Enable or disable auto-highlight mode. When enabled, action functions (`navigateSection`, `openCommand`, `clickElement`, `selectValue`, `fillFields`) automatically highlight the target element for 500ms before performing the action.

| Parameter | Type | Description |
|-----------|------|-------------|
| `on` | boolean | `true` to enable, `false` to disable |

**How it works**: each action highlights the element → waits 500ms (viewer reads) → removes highlight → performs the action. This prevents the highlight overlay from interfering with modals, dropdowns, or focus changes caused by the action.

**Search priority**: form elements (buttons, links, fields, grid rows) are searched first. Sections and commands are used as fallback only if the element is not found in the current form. This avoids false matches (e.g., "ОК" matching section "Покупки" via substring).

### `isHighlightMode()` → boolean

Check if auto-highlight mode is active.

### `highlight(text)`

Manually highlight a UI element by name (fuzzy match). Places a semi-transparent blue overlay (`rgba(0,100,255,0.25)`) with a border on the element. The overlay tracks element position via `requestAnimationFrame`.

| Parameter | Type | Description |
|-----------|------|-------------|
| `text` | string | Element name - button, link, field, group/panel, section, or command |

- Fuzzy match order: exact → startsWith → includes
- Search priority: popup items → commands → **form groups/panels** → form elements (buttons, fields) → sections
- Groups are matched by visible title or internal name (e.g., `highlight('Оргструктура')` finds the group panel)
- `pointer-events: none` - does not block clicks

### `unhighlight()`

Remove the highlight overlay.

## Example: Record a workflow with highlight, title slide, and captions

```js
await startRecording('recordings/create-order.mp4');

// Title slide with narration
await showTitleSlide('Создание заказа клиента', {
  subtitle: 'Демонстрация',
  speech: 'Создание заказа клиента. Демонстрация.'
});
await wait(1);
await hideTitleSlide();

// Presentation slide (optional)
await showImage('slides/overview.png', {
  speech: 'На этом слайде показана общая схема процесса'
});
await wait(1);
await hideImage();

setHighlight(true); // enable auto-highlight for all actions

// Steps: caption → pause → action (highlight is automatic)
await showCaption('Шаг 1. Переходим в раздел "Продажи"');
await wait(1.5);
await navigateSection('Продажи');

await showCaption('Шаг 2. Открываем заказы клиентов');
await wait(1.5);
await openCommand('Заказы клиентов');

await showCaption('Шаг 3. Создаем новый заказ');
await wait(1.5);
await clickElement('Создать');
await wait(2); // wait for form to load

await showCaption('Шаг 4. Заполняем шапку');
await wait(1.5);
await fillFields({ 'Организация': 'Конфетпром', 'Контрагент': 'Альфа' });
await wait(1);

await hideCaption();
setHighlight(false);
const result = await stopRecording();
console.log(`Recorded ${result.duration}s, ${(result.size / 1024 / 1024).toFixed(1)} MB`);
```

**Caption timing**: show the caption *before* the action - `showCaption` auto-waits for estimated TTS duration during recording. The subsequent `wait()` is absorbed by the credit system (no double-waiting). Add `wait()` *after* the action only when the next step needs the result to load (e.g., form opening).

**Highlight timing**: `setHighlight(true)` enables auto-mode - each action function highlights the target for 500ms, then removes the highlight before performing the action. No manual `highlight()`/`unhighlight()` calls needed. Enable after title slide, disable before `stopRecording()`.

## TTS Narration

Add voiceover to recorded videos. Captions shown via `showCaption()` are automatically collected during recording and can be synthesized into speech.

### Prerequisites

- **ffmpeg** - same as for video recording (ffprobe must be next to ffmpeg)
- **node-edge-tts** - `npm install --prefix tools/tts node-edge-tts` (for Edge TTS provider, free, no API key). Also works if installed globally or at project level - the resolver tries multiple locations automatically

### Configuration in `.v8-project.json`

```json
{
  "tts": {
    "provider": "edge",
    "voice": "ru-RU-DmitryNeural"
  }
}
```

For OpenAI-compatible provider:
```json
{
  "tts": {
    "provider": "openai",
    "apiKey": "sk-...",
    "voice": "alloy"
  }
}
```

For ElevenLabs:
```json
{
  "tts": {
    "provider": "elevenlabs",
    "apiKey": "sk_...",
    "voice": "JBFqnCBsd6RMkjVDRZzb"
  }
}
```
Note: `voice` is the ElevenLabs voice ID (not a name). Default model: `eleven_multilingual_v2` (supports Russian and other languages).

### `showCaption()` speech parameter

The `speech` option controls what text is narrated (vs displayed):

```js
await showCaption('Дт 60.02 - Кт 51');                           // narrates the displayed text
await showCaption('Дт 60.02 - Кт 51', { speech: 'Проводка: дебет шестьдесят ноль два, кредит пятьдесят один' }); // custom narration
await showCaption('Техническая информация', { speech: false });   // no narration for this caption
```

### `addNarration(videoPath, opts?)`

Generate TTS and merge audio with video. Call after `stopRecording()`.

**Parameters:**

| Parameter | Type | Description |
|-----------|------|-------------|
| `videoPath` | `string` | Path to the recorded MP4 file |
| `opts.captions` | `Array` | Explicit captions (default: from last recording or `.captions.json`). Each caption may include a `voice` field to override the global voice for that segment |
| `opts.provider` | `string` | `'edge'` (default), `'openai'`, or `'elevenlabs'` |
| `opts.voice` | `string` | Voice name (provider-specific) |
| `opts.apiKey` | `string` | API key (for openai) |
| `opts.apiUrl` | `string` | Endpoint (for openai) |
| `opts.model` | `string` | Model (for openai, default: `tts-1`) |
| `opts.ffmpegPath` | `string` | Path to ffmpeg binary |
| `opts.outputPath` | `string` | Output file (default: `video-narrated.mp4`) |

**Returns:** `{ file, duration, size, captions, warnings? }`

### `getCaptions()`

Returns captions from the current or last recording: `Array<{ text, speech, time, voice? }>`.

### Example: Record and narrate

```js
await startRecording('recordings/demo.mp4');
await showCaption('Переходим в раздел Банк и касса');
await wait(1.5);
await navigateSection('Банк и касса');
await showCaption('Открываем банковские выписки');
await wait(1.5);
await openCommand('Банковские выписки');
await hideCaption();
const video = await stopRecording();

// Add narration (reads tts config from .v8-project.json)
const narrated = await addNarration(video.file, { voice: 'ru-RU-DmitryNeural' });
console.log(`Narrated: ${narrated.file}, ${narrated.duration}s`);
```

### Re-narration

After recording, a `.captions.json` file is saved next to the video. You can re-narrate with a different voice without re-recording:

```js
const result = await addNarration('recordings/demo.mp4', { voice: 'ru-RU-SvetlanaNeural' });
```

## Troubleshooting

| Problem | Solution |
|---------|----------|
| "ffmpeg not found" | Install ffmpeg and ensure it's discoverable (see Prerequisites) |
| Recording file is 0 bytes | Check that output path is writable. ffmpeg may have crashed |
| Video is choppy | Add `wait()` between steps. Reduce `quality` for faster capture |
| "Already recording" | Call `stopRecording()` before starting a new recording |
| Recording stops on disconnect | Expected - auto-stop prevents orphaned ffmpeg processes |
| "No captions available" | Use `showCaption()` during recording, or pass `opts.captions` |
| TTS timeout | Check internet connection. Edge TTS requires network access |
| Audio cuts off between captions | Smart TTS wait should handle this automatically. If warnings appear, add longer `wait()` after `showCaption` |

---
name: 1c-web-test
description: Тестирование 1С через веб-клиент - автоматизация действий в браузере. Используй когда пользователь просит проверить, протестировать, автоматизировать действия в 1С через браузер
argument-hint: "сценарий на естественном языке"
allowed-tools:
  - Bash
  - Read
  - Write
  - Glob
  - Grep
---

# /web-test - Browser automation for 1C web client

Automates user interactions with 1C:Enterprise web client via Playwright - navigating sections, filling forms, reading tables and reports, filtering lists.

## Quick start

```bash
RUN="skills/1c-web-test/scripts/run.mjs"

# One-shot: opens browser → runs script → closes browser → exits
node $RUN run http://localhost:8081/bpdemo test-scenario.js

# Or pipe inline:
cat <<'SCRIPT' | node $RUN run http://localhost:8081/bpdemo -
await navigateSection('Продажи');
await openCommand('Заказы клиентов');
await clickElement('Создать');
await fillFields({ 'Клиент': 'Альфа' });
await clickElement('Провести и закрыть');
SCRIPT
```

## Setup (first time)

```bash
cd skills/1c-web-test/scripts && npm install
```

Requires Node.js 18+. `npm install` downloads Playwright and Chromium.

## URL resolution

Read `.v8-project.json` from project root. Each database has `id` and optional `webUrl`.
Construct URL as `http://localhost:8081/<id>` or use `webUrl` if set.
Use `/web-publish` first if the database is not published.

## Execution modes

### Autonomous mode (preferred for complete scenarios)

```bash
node $RUN run <url> script.js   # exits when done, no session
```

### Interactive mode (step-by-step development)

```bash
# 1. Start session (run_in_background=true, prints JSON when ready)
node $RUN start <url>

# 2. Execute scripts against running session
cat <<'SCRIPT' | node $RUN exec -
const form = await getFormState();
console.log(JSON.stringify(form, null, 2));
SCRIPT

# 2b. Execute without video recording (for debugging/testing)
cat script.js | node $RUN exec - --no-record

# 3. Screenshot
node $RUN shot result.png

# 4. Stop (logout + close)
node $RUN stop
```

`start` runs an HTTP server in background. Use `exec`/`shot`/`stop` from other shells.

### Writing exec scripts

All browser.mjs exports are globals - no `import` needed.
`console.log()` output is captured in the JSON response.
`writeFileSync` / `readFileSync` also available.

## API reference

### Navigation

#### `navigateSection(name)` → `{ navigated, sections, commands }`
Go to a top-level section (fuzzy match). Returns list of commands in that section.
```js
await navigateSection('Продажи');
// { navigated: 'Продажи', sections: [...], commands: ['Заказы клиентов', ...] }
```

#### `openCommand(name)` → form state
Open a command from the function panel (fuzzy). Returns form state of the opened form.
```js
const form = await openCommand('Заказы клиентов');
```

#### `navigateLink(url)` → form state
Open any 1C object by metadata path (Shift+F11 dialog). Bypasses section/command navigation.
```js
await navigateLink('Документ.ЗаказКлиента');
await navigateLink('РегистрНакопления.ЗаказыКлиентов');
await navigateLink('Справочник.Контрагенты');
```

#### `openFile(path)` → form state
Open an external data processor or report (EPF/ERF) via File → Open. Handles the security confirmation dialog automatically.
```js
const form = await openFile('C:\\WS\\build\\МояОбработка.epf');
const form = await openFile('build/МояОбработка.epf'); // relative paths work too
```

#### `switchTab(name)` → form state
Switch to an already-open tab/window (fuzzy match).

### Reading form state

#### `getFormState()` → `{ form, formCount, openForms, fields, buttons, tabs, navigation?, table, tables, filters, reportSettings? }`
Returns current form structure. This is the primary way to understand what's on screen.

**form** - active form number, or `null` when no form is open (desktop).

**formCount** - number of open forms. Use this to know how many windows are stacked. `0` means desktop.

**openForms** - array of all open form numbers (e.g. `[0, 1]`). Works even when the open-windows tab bar is hidden in 1C settings.

**modal** - `true` when the active form is a modal dialog blocking the UI. Only present when modal is active.

**openTabs** - array of `{ name, active? }` from the open-windows tab bar. Only present when the tab bar is enabled in 1C settings. Do NOT rely on this - use `formCount`/`openForms` instead.

**fields** - each field has: `name`, `value`, `label?`, `actions?` (select, clear, open), `required?` (true for unfilled mandatory fields)

**navigation** - form navigation panel links (for objects with subordinate catalogs): `[{ name, active? }]`. Clickable via `clickElement()`. Only present when the form has a navigation panel (e.g. "Основное", "Объекты метаданных", "Подсистемы").

**tables** - array of all visible grids: `[{ name, columns, rowCount, label? }]`. `label` is the visual group title shown on screen (e.g. "Входящие"), absent when grid has no visible title. Use `readTable()` for actual data.

**table** - backward-compatible alias for the first grid: `{ present, columns, rowCount }`.

**reportSettings** - for DCS reports: human-readable filter settings instead of raw technical names:
```js
const form = await getFormState();
// form.reportSettings = [
//   { name: "Склад", enabled: true, value: "Склад бытовой техники", actions: ["select"] },
//   { name: "Номенклатура", enabled: false, value: "" }
// ]
```

**errorModal** - if present, 1C showed an error dialog. Read the message and decide how to proceed.

**confirmation** - if present, a Yes/No dialog is shown. Call `clickElement('Да')` or `clickElement('Нет')`.

**errors.stateText** - array of SpreadsheetDocument state messages (e.g. `"Не установлено значение параметра \"X\""`, `"Отчет не сформирован..."`, `"Изменились настройки..."`). Present when the report area shows an info bar instead of data.

### Reading data

#### `readTable({ maxRows?, offset?, table? })` → `{ columns, rows, total, shown, offset }`
Read actual grid data with pagination. Each row is `{ columnName: value }`.

| Option | Default | Description |
|--------|---------|-------------|
| `maxRows` | 20 | Max rows to return per call |
| `offset` | 0 | Skip first N rows |
| `table` | - | Grid name from `tables[]` (for multi-grid forms) |

Special row fields:
- `_kind: 'group'` - hierarchical group row
- `_kind: 'parent'` - parent row in hierarchy
- `_tree: 'expanded'|'collapsed'` - tree node state
- `_level: N` - nesting depth in tree view
- `_selected: true` - row is selected (highlighted). Use with `clickElement({ modifier: 'ctrl'|'shift' })` to verify multi-selection
- `hierarchical: true` - list has groups (on result object)
- `viewMode: 'tree'` - tree view active (on result object)

```js
const t = await readTable({ maxRows: 50 });
console.log('Columns:', t.columns);
console.log('Rows:', t.rows.length, 'of', t.total);
// Pagination:
const page2 = await readTable({ maxRows: 50, offset: 50 });
```

#### `readSpreadsheet()` → `{ title?, headers?, data?, totals?, rows?, total }`
Read report output (SpreadsheetDocument) after clicking "Сформировать".

Returns structured data when header row is detected:
```js
await clickElement('Сформировать');
await wait(5);
const report = await readSpreadsheet();
// { title: "Остатки товаров", headers: ["Номенклатура", "Склад", "Количество"],
//   data: [{ "Номенклатура": "Бумага", "Склад": "Основной", "Количество": "150" }, ...],
//   totals: { "Количество": "1250" }, total: 42 }
```

Falls back to `{ rows: string[][], total }` when headers can't be detected.

#### `getSections()` → `{ activeSection, sections, commands }`
Read section panel and commands without navigating.

#### `getCommands()` → `string[]`
Commands of the current section.

#### `getPageState()` → `{ activeSection, activeTab, sections, tabs }`
Sections + all open tabs.

### Actions

#### `clickElement(text, { dblclick?, table?, expand?, modifier? })` → form state
Click button, hyperlink, tab, navigation panel link, or grid row (fuzzy match).

- `table` - scope button search to a specific grid's command panel (by name from `tables[]`):
  ```js
  await clickElement('Добавить', { table: 'Исходящие' }); // clicks "Добавить" near "Исходящие" grid
  ```
- Single click selects a row in a list. **Double-click opens** the item:
  ```js
  await clickElement('0000-000227', { dblclick: true }); // opens document
  ```
- Returns `submenu[]` when a menu opens - click again with item name:
  ```js
  const r = await clickElement('Еще');
  // r.submenu = ['Расширенный поиск', 'Настройки', ...]
  await clickElement('Расширенный поиск');
  ```
- **Tree nodes**: default click = **select** (highlight row). Use `{ expand: true }` to **expand/collapse**:
  ```js
  await clickElement('ИСУ ФХД');                      // select row
  await clickElement('ИСУ ФХД', { expand: true });    // expand/collapse
  ```
- **Multi-select rows** with `modifier: 'ctrl'` (add to selection) or `modifier: 'shift'` (select range):
  ```js
  await clickElement('Номенклатура 1');                          // select first row
  await clickElement('Номенклатура 2', { modifier: 'ctrl' });   // add to selection
  await clickElement('Номенклатура 5', { modifier: 'shift' });  // select range 2..5
  // Verify selection:
  const t = await readTable();
  t.rows.filter(r => r._selected);  // rows with _selected: true
  ```
- **SpreadsheetDocument cells** (report drill-down): first argument can be `{ row, column }` object to click a cell in a rendered report. Coordinates match `readSpreadsheet()` output:
  ```js
  const report = await readSpreadsheet();
  // report.data[0] = { 'К1': 'Материалы строительные', 'К6': '150 000', ... }

  // By data row index + column header name
  await clickElement({ row: 0, column: 'К6' }, { dblclick: true });

  // By cell value filter (fuzzy match)
  await clickElement({ row: { 'К1': 'Материалы' }, column: 'К6' }, { dblclick: true });

  // Totals row
  await clickElement({ row: 'totals', column: 'К6' }, { dblclick: true });
  ```
  Text search also works as fallback - searches inside spreadsheet iframes:
  ```js
  await clickElement('150 000', { dblclick: true }); // finds cell by text in report
  ```

#### `fillFields({ name: value })` → `{ filled, form }`
Fill form fields by label (fuzzy match). Auto-detects field type.

| Value | Field type | Method |
|-------|-----------|--------|
| `'Конфетпром'` | Reference | Clipboard paste + typeahead |
| `'5000'` | Plain text | Clipboard paste |
| `'true'` / `'да'` | Checkbox | Toggle |
| `'Оплата поставщику'` | Radio | Fuzzy label match |
| `''` / `null` | Any (except checkbox/radio) | Clear via Shift+F4 |

**DCS report filters**: use human-readable label names. Checkbox is auto-enabled:
```js
await fillFields({
  'Склад': 'Склад бытовой техники',   // auto-enables "Склад" checkbox + fills value
  'Номенклатура': 'Вентилятор'          // same: enables checkbox + fills
});
```

Returns `{ filled: [{ field, ok, value, method }], form: {...} }`.
Method is one of: `'clear'` | `'toggle'` | `'radio'` | `'paste'` | `'dropdown'` | `'form'` | `'typeahead'`

#### `selectValue(field, search, opts?)` → form state with `selected`
Select a value from reference field via dropdown or selection form. More reliable than `fillFields` for reference fields that need exact selection from a catalog. Pass empty `search` (`''` or `null`) to clear the field (Shift+F4).

`search` - string for simple search, or `{ field: value }` object for per-field advanced search:
```js
await selectValue('Организация', 'Конфетпром');
// result.selected = { field: 'Организация', search: 'Конфетпром', method: 'dropdown'|'form' }

// Per-field search (disambiguate by multiple columns):
await selectValue('Документ', { 'Номер': '0000-000601', 'Дата': '29.12.2016' }, { type: 'Реализация (акт' });
```

For **composite-type fields** (accepting multiple types), specify `type` to first select the type, then the value:
```js
await selectValue('Документ', '0000-000601', { type: 'Реализация (акт' });
// Clears field → opens type dialog → picks type via Ctrl+F → picks value from selection form
// result.selected = { field: 'Документ', search: '0000-000601', type: 'Реализация (акт', method: 'form' }
```

Also supports DCS labels - auto-enables the paired checkbox.

#### `fillTableRow(fields, opts)` → form state
Fill table row cells via Tab navigation. Value is a plain string, `{ value, type }` for composite-type cells, or `''`/`null` to clear (Shift+F4).

| Option | Description |
|--------|-------------|
| `tab` | Switch to tab before filling |
| `add` | Add new row before filling |
| `row` | Edit existing row by 0-based index |
| `table` | Grid name from `tables[]` (for multi-grid forms) |

```js
// Add new row:
await fillTableRow(
  { 'Номенклатура': 'Бумага', 'Количество': '10', 'Цена': '100' },
  { tab: 'Товары', add: true }
);
// Edit existing row:
await fillTableRow(
  { 'Количество': '20' },
  { tab: 'Товары', row: 0 }
);
// Multi-grid form - add row to specific table:
await fillTableRow(
  { 'Объект': 'БДДС' },
  { table: 'Исходящие', add: true }
);
// Composite-type cell (e.g. SubConto accepting multiple types):
await fillTableRow(
  { 'СубконтоКт1': { value: 'Голованов', type: 'Физическое лицо' } },
  { tab: 'Проводки' }
);
```

- Tab-based sequential navigation - field order set by 1C form config
- Fuzzy cell match: "Количество" matches "ТоварыКоличество"
- Reference cells auto-detected by autocomplete popup

#### `deleteTableRow(row, { tab?, table? })` → form state
Delete row by 0-based index. `table` targets a specific grid on multi-grid forms.

#### `closeForm({ save? })` → form state with `closed`
Close the current form via Escape. Returns form state with `closed: true/false` indicating whether the form actually closed.

| Argument | Behavior |
|----------|----------|
| `{ save: false }` | Auto-clicks "Нет" on confirmation |
| `{ save: true }` | Auto-clicks "Да" on confirmation |
| `{}` (omitted) | Returns `confirmation` field if dialog appears |

**`closed`** - `true` if the form was closed (form number changed), `false` if it stayed open (e.g. Escape was ignored). Always check this to confirm the form actually closed. After closing, check `formCount` to see how many forms remain.

Preferred over `clickElement('×')` - close buttons on tabs are ambiguous.

#### `filterList(text, opts?)` → form state
Filter list. Simple mode searches all columns, advanced mode targets a specific field.

```js
await filterList('КП00-000018');                          // simple - all columns
await filterList('Мишка', { field: 'Наименование' });     // advanced - specific column
await filterList('Мишка', { field: 'Наименование', exact: true }); // exact match
```

Works on hierarchical catalogs too (flattens the view).

#### `unfilterList({ field? })` → form state
Clear filters. Without arguments clears all, with `{ field }` clears specific badge.

### Utility

#### `screenshot()` → PNG Buffer
#### `wait(seconds)` → form state
#### `getPage()` → Playwright Page (raw, for advanced scripting)
#### `startRecording(path, opts?)` / `stopRecording()` → MP4 video recording (`{ force: true }` to restart if already recording)
#### `showCaption(text, opts?)` / `hideCaption()` → text overlay on page
#### `showTitleSlide(text, opts?)` / `hideTitleSlide()` → full-screen title card (intro/outro)
#### `isRecording()` → boolean
#### `setHighlight(on)` / `isHighlightMode()` → auto-highlight mode for video
#### `highlight(text)` / `unhighlight()` → manual element highlighting (error lists available elements)
#### `addNarration(videoPath, opts?)` → narrated MP4 with TTS voiceover
#### `getCaptions()` → caption timestamps from last recording

See [recording.md](recording.md) for setup (ffmpeg), highlight mode, TTS narration, API details, and examples.
If `.v8-project.json` has `ffmpegPath`, pass it to `startRecording({ ffmpegPath })`.
If `.v8-project.json` has `tts` config, pass it to `addNarration()` (provider, voice, apiKey).

## Common patterns

### Create and save a document

```js
await navigateSection('Продажи');
await openCommand('Заказы клиентов');
await clickElement('Создать');
await fillFields({ 'Организация': 'Конфетпром', 'Контрагент': 'Альфа' });
await fillTableRow({ 'Номенклатура': 'Бумага', 'Количество': '10' }, { tab: 'Товары', add: true });
await clickElement('Провести и закрыть');
```

### Open item from list

```js
await clickElement('КП00-000227', { dblclick: true });
// Always use { dblclick: true } - single click only selects the row
```

### Work with hierarchical lists

```js
await filterList('Конфетпром');                               // flatten + search
await clickElement('Конфетпром ООО', { dblclick: true });     // open
await closeForm();
await unfilterList();                                          // restore hierarchy
```

### Generate and read a report

```js
// Fill report filters using readable labels
await fillFields({ 'Склад': 'Основной склад' });
await clickElement('Сформировать');
await wait(5);
const report = await readSpreadsheet();
console.log('Title:', report.title);
console.log('Data rows:', report.data?.length);
```

### Drill-down report cells

```js
// Generate report
await clickElement('Сформировать');
await wait(5);
const report = await readSpreadsheet();

// Double-click cell to open drill-down (uses coordinates from readSpreadsheet)
await clickElement({ row: 0, column: 'К6' }, { dblclick: true });
// Modal dialog "Выбор поля" opens
await clickElement('Регистратор');
await clickElement('Выбрать');
await wait(10);
const drilldown = await readSpreadsheet();
```

### Work with multi-grid forms

Some forms have multiple grids (e.g. "Входящие" and "Исходящие" tables on a single form). Without `table`, buttons like "Добавить" hit the first match and `readTable` reads the first grid - which may not be the one you need.

**Step 1 - discover tables** via `getFormState()`:
```js
const form = await getFormState();
// form.tables = [
//   { name: "ДеревоБизнесПроцессов", columns: ["Полный код", "Бизнес-процесс"], rowCount: 21 },
//   { name: "Входящие", label: "Входящие", columns: ["Объект", "Бизнес-процесс источник", ...], rowCount: 1 },
//   { name: "Исходящие", label: "Исходящие", columns: ["Объект", "Бизнес-процесс приемник", ...], rowCount: 1 }
// ]
```

**Step 2 - use `table` name** in any grid operation:
```js
// Read specific table
const t = await readTable({ table: 'Исходящие' });

// Add row - fillTableRow with add:true already clicks the right "Добавить" button
await fillTableRow({ 'Объект': 'БДДС' }, { table: 'Исходящие', add: true });

// Or click buttons separately
await clickElement('Добавить', { table: 'Входящие' });

// Delete from specific table
await deleteTableRow(0, { table: 'Исходящие' });
```

Table matching accepts both technical name (`tables[].name`) and visual label (`tables[].label`). Label is the group title shown on screen - useful when working from screenshots. Name match takes priority over label match.

### Keyboard shortcuts

```js
const page = await getPage();
await page.keyboard.press('F8');  // example: create new item in focused reference field
```

| Key | Context | Action |
|-----|---------|--------|
| `F8` | Reference field focused | Create new catalog item |
| `Shift+F4` | Any input field focused | Clear field value (auto via `''`/`null` in fillFields/selectValue/fillTableRow) |
| `F4` | Reference field focused | Open selection form |
| `Alt+F` | List/table form | Open advanced search dialog |

### Closing forms - which method to use

| Goal | Method |
|------|--------|
| Post & close document | `clickElement('Провести и закрыть')` |
| Save & close catalog | `clickElement('Записать и закрыть')` |
| Close without saving | `closeForm({ save: false })` |
| Close and save | `closeForm({ save: true })` |
| Close (manual confirm) | `closeForm()` - returns `confirmation` if dialog appears |

## Exec response format

```json
{ "ok": true, "output": "...console.log output...", "elapsed": 3.2 }
```

On error (auto-screenshot taken):
```json
{ "ok": false, "error": "Element not found", "output": "...", "screenshot": "error-shot.png", "elapsed": 1.5 }
```

## Avoiding loops

- **Max 2 attempts per operation.** If an action fails twice with the same approach - stop and report to the user
- **Not found = not found.** If `filterList` returns 0 rows or `readTable` is empty after filtering - the item likely doesn't exist in this database. Don't retry the same search 5 times with slight variations
- **Try a different approach, not the same one.** Couldn't find via section navigation? Try `navigateLink`. Couldn't find via simple search? Try advanced search with a specific field. But don't repeat the same method
- **Report partial results.** If you found the list but not the specific item - say so. Show what IS available instead of silently retrying

## Important notes

- **Headed mode** - 1C requires visible browser, no headless
- **Startup time** - 1C loads 30-60s on initial connect (built into `start`)
- **Fuzzy matching** - all name lookups: exact > startsWith > includes
- **Clipboard paste** - all text fields filled via Ctrl+V (triggers 1C events properly)
- **Cyrillic in bash** - use `cat <<'SCRIPT' | node $RUN exec -` to avoid escaping issues
- **Non-breaking spaces** - 1C uses `\u00a0` instead of regular spaces. All matching is normalized internally
- **Section panel display** - `navigateSection()` works with any panel position (side, top) but requires "Picture and text" or "Text" display mode. Icon-only mode is not supported - API cannot read section names from icons alone