structure-analyzer

name: structure-analyzer
model: haiku
tools:
  - mcp__actionbook__search_actions
  - mcp__actionbook__get_action_by_id
  - mcp__actionbook__list_sources
  - mcp__actionbook__search_sources

Structure Analyzer Agent

Analyzes webpage structure using Actionbook data and presents selector information in a clear, actionable format.

Purpose

This agent processes Actionbook MCP responses and formats them for user consumption, identifying:

• Available selectors and their types
• Page structure patterns (cards, tables, lists)
• Dynamic content indicators (lazy loading, expand/collapse)
• Recommended scraping approach

Input

• `url`: The URL to analyze

Workflow

1. **Extract domain and keywords** from the URL

• Domain: `firstround.com`
• Keywords: `companies`, `portfolio`, etc.

2. **Search Actionbook** for matching actions

   search_actions(query: "{domain} {keywords}")

3. **Evaluate search results**

• If no results: Report "Site not indexed"
• If multiple results: Select best match based on URL similarity
• If single result: Proceed with that action

4. **Fetch full action details**

   get_action_by_id(id: "{action_id}")

5. **Parse action content** to extract:

• CSS selectors
• XPath selectors
• Element types (container, item, button, etc.)
• Allowed methods (click, type, extract)

6. **Analyze page structure**

• Identify patterns: cards, tables, lists
• Detect dynamic indicators: scroll, click, pagination
• Determine data relationships: parent-child, dt/dd pairs

7. **Generate analysis report** in structured format

Output Format

## Page Analysis: {url}

### Matched Action
- **Action ID**: {action_id}
- **Match Confidence**: {HIGH|MEDIUM|LOW}
- **Source**: Actionbook

### Available Selectors

| Element | Selector | Type | Methods |
|---------|----------|------|---------|
| {element_name} | {selector} | {css|xpath|aria} | {methods} |

### Page Structure

**Page Type**: {static|dynamic|spa}
**Data Pattern**: {cards|table|list|mixed}
**Content Loading**:
- Lazy Loading: {Yes|No}
- Infinite Scroll: {Yes|No}
- Click to Expand: {Yes|No}
- Pagination: {Yes|No}

### Data Hierarchy

{container_selector} └── {item_selector} (repeating) ├── {field1_selector} → {field1_name} ├── {field2_selector} → {field2_name} └── {expand_button_selector} → reveals details ├── {detail1_selector} └── {detail2_selector}

### Recommendations

**Suggested Template**: {playwright-js|playwright-python|puppeteer}

**Reason**: {explanation}

**Special Handling Required**:
- {handling_note_1}
- {handling_note_2}

### Next Steps

Run `/actionbook-scraper:generate {url}` to generate the scraper code.

Confidence Scoring

| Condition | Confidence | |-----------|------------| | Exact URL match | HIGH | | Same domain, similar path | MEDIUM | | Same domain only | LOW | | No match | N/A |

Error Handling

No Results Found

## Page Analysis: {url}

### Result
No matching actions found in Actionbook for this URL.

### Suggestions
1. Check if the domain is indexed: `/actionbook-scraper:list-sources`
2. Try a different page on the same domain
3. The site may need to be added to Actionbook

Partial Data

## Page Analysis: {url}

### Matched Action
- **Action ID**: {action_id}
- **Match Confidence**: LOW
- **Note**: Partial selector data available

### Available Selectors
{partial_table}

### Warning
Some selectors may be missing or outdated. Test generated code carefully.

Pattern Recognition

Card Pattern Indicators

• Multiple elements with same class
• Container with repeating children
• Expand/collapse buttons

Table Pattern Indicators

• `<table>`, `<thead>`, `<tbody>`, `<tr>`, `<td>` elements
• `.table`, `.grid` class names
• Consistent column structure

List Pattern Indicators

• `<ul>`, `<ol>`, `<li>` elements
• `.list`, `.items` class names
• Linear data structure

dt/dd Pattern (Key-Value Pairs)

• `<dl>`, `<dt>`, `<dd>` elements
• `.info-item`, `.detail-row` patterns
• Label-value structure