Performance Testing with Browser Metrics

Performance bugs are invisible until they're not. A page loads in 500ms on your machine but takes 8 seconds on a real user's connection. A memory leak crashes the tab after 20 minutes of use. Certyn's get_performance_metrics tool gives AI agents access to the same metrics Chrome tracks internally.

How It Works

The agent enables Chrome's Performance domain, collects a snapshot of all tracked metrics, then disables it. The result is a set of key numbers that reveal how the page is actually performing — not how it looks, but how it behaves.

Key Metrics

Page load

• Documents — number of documents (frames) in the page
• Nodes — total DOM node count (high counts = slow rendering)
• FirstContentfulPaint — time until the user sees meaningful content

Rendering

• LayoutCount — how many times the browser recalculated layout
• LayoutDuration — total time spent on layout (in seconds)
• RecalcStyleCount — number of style recalculations
• RecalcStyleDuration — total time spent recalculating styles

JavaScript

• ScriptDuration — total time spent executing JavaScript
• TaskDuration — total time spent on all browser tasks

Memory

• JSHeapUsedSize — current JavaScript heap usage (MB)
• JSHeapTotalSize — total allocated heap (MB)

What to Test

Page load performance

Collect metrics after navigating to a page. Check:

• Is FirstContentfulPaint under 1.8 seconds?
• Is the DOM node count reasonable (under 1,500 for most pages)?
• Is ScriptDuration under 1 second?

Interaction performance

Collect metrics before and after an interaction. Compare:

• Did LayoutCount spike? (layout thrashing)
• Did JSHeapUsedSize grow significantly? (potential memory leak)
• Is TaskDuration reasonable for the action?

Memory leaks

Repeat an action multiple times and check JSHeapUsedSize after each:

• Open and close a modal 10 times — does memory grow?
• Navigate between pages — does the heap keep increasing?
• Add and remove list items — is memory reclaimed?

Example Scenarios

Baseline page performance

"Navigate to the dashboard and get performance metrics. Check if FirstContentfulPaint is under 2 seconds and DOM node count is under 1500."

Test after heavy interaction

"Open the data table, sort by each column, apply 3 filters, then get performance metrics. Check if LayoutCount is reasonable and JSHeapUsedSize hasn't grown excessively."

Slow network performance

Combine with simulate_network for realistic conditions:

"Simulate a 3G network, navigate to the homepage, wait for it to fully load, then get performance metrics. Is FirstContentfulPaint under 5 seconds on 3G?"

Memory leak detection

"Navigate to the chat page. Get performance metrics (baseline). Send 50 messages. Get performance metrics again. Compare JSHeapUsedSize — it should not have grown by more than 20%."

Tips

• Collect metrics at natural checkpoints: after page load, after user actions, after idle periods
• Compare metrics across runs rather than relying on absolute thresholds
• Combine with network simulation to test under realistic conditions
• High LayoutCount relative to user actions usually indicates layout thrashing — a common performance problem
• DOM node counts over 1,500 can slow down every interaction on the page