frago

Here’s a production-ready frago Recipe for automating competitor pricing analysis across 3 e-commerce sites. The workflow includes:

**Recipe Metadata:**
```yaml
name: competitor_price_tracker
version: 1.0.0
steps:
  - id: fetch_sitemap
    action: frago.http.get
    input:
      url: "https://{competitor_domain}/sitemap.xml"
      headers:
        User-Agent: "Mozilla/5.0 (frago-agent/1.0)"
    output: sitemap_data
    error_handler:
      retry: 3
      on_failure: notify_ops_team

  - id: scrape_product_pages
    action: frago.browser.navigate
    input:
      url: "{{ steps.fetch_sitemap.outputs.sitemap_data.urls.product }}"
      cdp_commands:
        - "Page.navigate"
        - "Runtime.evaluate"
    output: product_data
    error_handler:
      screenshot: true
      on_failure: log_error_and_continue

  - id: compare_prices
    action: frago.data.compare
    input:
      reference_data: "{{ steps.scrape_product_pages.outputs.product_data }}"
      threshold: 0.15
    output: price_alerts
```

**Persistent Run Context:**
- The Recipe stores outputs in frago’s Run context (`frago run get-context --id {run_id}`), including:
  - Raw HTML from Step 1
  - Screenshots of failed pages (Step 2)
  - JSON array of price alerts (Step 3)

**Execution & Monitoring:**
```bash
# Start the workflow
frago run execute --recipe competitor_price_tracker.yaml --env PROD

# Monitor in real-time
frago run watch --id {run_id} --output json

# Retrieve results
frago run get-context --id {run_id} | jq '.price_alerts[] | select(.severity == "high")'
```

**Output:**
```json
{
  "price_alerts": [
    {
      "product_id": "SKU-12345",
      "competitor": "AcmeCorp",
      "our_price": 99.99,
      "competitor_price": 84.99,
      "savings_pct": 15.01,
      "severity": "high",
      "screenshot_url": "s3://frago-runs/{run_id}/screenshots/SKU-12345.png"
    }
  ]
}
```

**Key Features Demonstrated:**
1. **Metadata-Driven:** The Recipe is defined in YAML with explicit steps, inputs, and error handling.
2. **Multi-Runtime:** Combines HTTP requests (Step 1), CDP browser control (Step 2), and data comparison (Step 3).
3. **Persistent Context:** All outputs are stored for auditing or downstream processing.
4. **Error Resilience:** Failed steps trigger retries or notifications without halting the entire workflow.