Browser metrics

The JSON examples below show selected fields from the full response.

These are page-wide signals, not React-only measurements. Repeat the same browser or device flow for every before-and-after sample.

browser_get_memory

Reads the current page-wide JavaScript heap estimate exposed by Chromium.

Input: No fields; pass {}.

npx @genie-react/cli call browser_get_memory '{}' \
  --session logical-review --json

Output (selected fields):

{
  "supported": true,
  "usedJSHeapSize": 24576000,
  "totalJSHeapSize": 36700160,
  "jsHeapSizeLimit": 4294705152,
  "note": "Browser JavaScript (V8) heap for the whole page, not React-specific memory. Values are coarsened by the browser for security."
}

Agent use: Record usedJSHeapSize before and after the same repeated flow; if either sample is unsupported or from a different document, do not claim a memory improvement.

Limits: This is an instantaneous, coarsened reading from non-standard Chromium performance.memory. It covers the whole page, not one React component. Firefox, Safari, React Native, and runtimes without numeric heap fields return supported:false with a note; absent fields are not zero. The call creates no subscription and needs no cleanup.

browser_measure_memory

Requests a page-wide memory estimate with per-realm attribution when the browser supports it.

Input: No fields; pass {}.

npx @genie-react/cli call browser_measure_memory '{}' \
  --session logical-review --json

Output (selected fields):

{
  "supported": true,
  "bytes": 28400000,
  "breakdown": [
    {
      "bytes": 21000000,
      "attribution": [
        { "url": "https://app.test/checkout", "scope": "Window" }
      ],
      "types": ["JavaScript", "DOM"]
    }
  ],
  "note": "Page-wide browser memory estimate across all realms (JS heap, DOM, etc.), not React-specific memory. The browser may delay sampling."
}

Agent use: Compare bytes and the same attribution scopes across repeated, controlled flows; when supported:false, fall back to browser_get_memory and label the weaker evidence.

Limits: performance.measureUserAgentSpecificMemory() requires a Chromium-based browser and a cross-origin-isolated page with COOP same-origin and COEP require-corp. Sampling can be delayed and can return a tool error if the browser rejects it. The total includes all page realms, JS heap, and DOM memory, not one component. The call leaves no ongoing sampler.

browser_fps

Samples page-wide animation-frame health for one fixed time window.

Input: durationMs is an optional integer from 250–10,000 and defaults to 5,000.

npx @genie-react/cli call browser_fps '{"durationMs":5000}' \
  --session logical-review --json

Output (selected fields):

{
  "durationMs": 5004,
  "frames": 299,
  "avgFps": 59.8,
  "worstFrameMs": 34.2,
  "longFrames": 0,
  "droppedFrames": 1,
  "refreshRate": 60,
  "hidden": false,
  "visibility": {
    "startedHidden": false,
    "endedHidden": false,
    "hiddenDuringSample": false
  },
  "throttleState": "none-detected",
  "calibration": {
    "method": "frame-interval-distribution",
    "confidence": "high",
    "sampleCount": 299,
    "intervalMs": { "p10": 16.6, "p50": 16.7, "p90": 16.8 },
    "modesHz": [60]
  },
  "refreshRateModes": { "firstHalfHz": 60, "secondHalfHz": 60 },
  "comparable": true,
  "notComparableReasons": [],
  "verdict": "smooth"
}

Agent use: Drive the same scroll, animation, or interaction during each sample and compare only runs with comparable:true; otherwise fix visibility or sampling conditions and repeat the run.

Limits: The call blocks for its sampling window, so start or drive the interaction from browser automation before or concurrently with it. It measures page-wide main-thread frame pacing, not React work alone. A tab hidden at start errors. Hiding it during sampling, sparse calibration, or a refresh-rate mode change sets comparability evidence that must be read. The internal timeout ends a stalled sample; no animation-frame subscription remains after the result.