measure-cwv

name	measure-cwv
description	Measure Core Web Vitals (LCP, INP, CLS) on a live page using PerformanceObserver injection. Returns metrics + thresholds verdict per web.dev good/needs-improvement/poor bands. Chromium-only — Firefox and WebKit ship partial PerformanceObserver coverage.

/measure-cwv

Single-backend skill. Calls measure_cwv on the rolepod-uiproof MCP server. No fallback (D-024).

When to use

• Verifying that a page meets the Core Web Vitals targets for a UX change.
• Comparing LCP / INP / CLS before vs after a UI change.
• Establishing a CWV baseline for a landing page.
• Running a CWV check inside a Verify-phase flow alongside /audit-a11y and /check-errors.

When NOT to use

• Backend latency or server-side p95/p99 timing — that is the parent rolepod's performance-engineer agent territory.
• Bundle size / build-output inspection — same.
• Synthetic load tests (k6, Locust) — same.
• Firefox or WebKit targets — the underlying PerformanceObserver entry types are chromium-specific; the tool refuses other browsers with cwv_unsupported_browser.

Inputs

• url — URL to measure.
• viewport — optional { width, height }.
• emulate.network_throttle — one of offline | slow-3g | fast-3g | slow-4g | fast-4g | no-throttling.
• emulate.cpu_throttle — 1 (none) to 20 (very slow). Chromium only.
• observe_ms — observation window after load. Default 5000ms.
• interactions[] — optional steps to elicit INP samples. Each step is { kind: "click", query }, { kind: "type", query, text }, { kind: "key", key }, or { kind: "scroll", direction, amount? }.
• thresholds — overrides for lcp_ms (2500), inp_ms (200), cls (0.1).

Outputs

• run_id — folder under the run root (see Evidence routing).
• metrics — { lcp_ms, inp_ms, cls }.
• verdict — per-metric good | needs-improvement | poor | unmeasured. INP is unmeasured when no interactions were driven.
• status — pass (all good) | warn (any needs-improvement) | fail (any poor).
• thresholds — the effective thresholds used.
• report_path — path to cwv.json with full samples.

Process

1. Build measure_cwv input from the user's intent (URL, optional throttle, optional interactions).
2. Call the tool.
3. Surface the three metrics + verdict inline. Reference the report path for the full sample list.

Evidence routing

Run artifacts are saved under:

• Standalone: .rolepod-uiproof/artifacts/measure_cwv_<ts>_<uuid>/
• With rolepod parent (detected via the marker file <git-root>/.rolepod/parent-active written by the parent's SessionStart hook): <git-root>/.rolepod/evidence/<ts>-rolepod-uiproof-measure-cwv/

Either way the run directory contains a manifest.json per Extension Protocol v1 so the parent's check-work skill can aggregate the result into the verify report. Standalone users can read the manifest themselves — same shape.

If the tool is unavailable

Surface plainly:

The /measure-cwv skill needs the rolepod-uiproof MCP server, which is not currently available. Confirm the plugin is installed and try again.

Do not attempt another backend (D-024).

Examples

Plain LCP / CLS check on a landing page

{
  "url": "https://example.com",
  "observe_ms": 5000
}

INP will be reported as unmeasured — no interactions were driven.

Eliciting an INP sample after a click

{
  "url": "https://example.com/app",
  "observe_ms": 3000,
  "interactions": [
    { "kind": "click", "query": "Open menu" }
  ]
}

Throttled measurement (slow 4G, 4x CPU)

{
  "url": "https://example.com",
  "emulate": { "network_throttle": "slow-4g", "cpu_throttle": 4 },
  "observe_ms": 7000
}