speedprompt.ai

PageSpeed · Findings → Agent Prompts

Run a Lighthouse scan, turn the rough edges into concrete findings, and hand your agent a prompt that already has the context, evidence, and likely files to inspect.

$ scan

Try this example, or replace it with your own URL.

◊ Free · 20 Scans / 30 Min ◊ No Sign-In Required ◊ Results in About 15s

01 Step

Scan the real page.

We run PageSpeed Insights against the live URL you care about for both Mobile and Desktop.

02 Step

Review every finding.

Each failing audit gets packaged with severity, evidence, scores, and a prompt-ready summary.

03 Step

Hand it to your agent.

Copy the prompt, share the result, or download the skills bundle and move straight into execution.

Sample Output

Prompts your agent actually knows what to do with.

Curated templates for the Lighthouse audits teams fix most often. Each one includes evidence, acceptance criteria, and the likely files to inspect first.

sample scan findings

01 Eliminate Render-Blocking Resources CSS and JavaScript files in the head block the browser from rendering content until they are fully downloaded and parsed. high render-blocking-resources

prompt.md static embedded prompt example

# Eliminate Render-Blocking Resources

CSS and JavaScript files in the `<head>` block the browser from rendering any content until they are fully downloaded and parsed. This directly delays First Contentful Paint (FCP) and Largest Contentful Paint (LCP).

## What to fix
- **CSS:** Extract critical above-the-fold CSS and inline it in a `<style>` tag. Load the full stylesheet asynchronously using `<link rel="preload" as="style" onload="this.rel='stylesheet'">`.
- **JavaScript:** Add `defer` or `async` attributes to non-critical `<script>` tags. Move third-party scripts (analytics, chat widgets) to `defer` or load them after `DOMContentLoaded`.
- **Fonts:** Use `font-display: swap` and preload critical font files with `<link rel="preload" as="font" crossorigin>`.

## How to verify
Check the Lighthouse filmstrip - first paint should occur before all external resources finish loading.

02 Serve Correctly Sized Images Images are being served at a larger resolution than their display size requires, wasting bandwidth on smaller viewports. medium uses-responsive-images

prompt.md static embedded prompt example

# Serve Correctly Sized Images

Images are being served at a larger resolution than their display size requires. A 2000px image displayed at 400px wastes bandwidth.

## What to fix
- **Use `srcset` and `sizes` attributes** to serve appropriately sized images for each viewport.
- **Generate multiple image sizes** at build time (e.g., 320w, 640w, 1024w, 1920w).
- **Use an image CDN** that auto-resizes based on request parameters.
- **Set the `sizes` attribute accurately** to tell the browser how wide the image will be displayed.

## How to verify
In DevTools, compare the natural image dimensions to the rendered size - they should be within 2x for retina.

What We Do

Built for teams that already know the problem is not finding bugs. It is shipping the fix.

speedprompt.ai sits between scan output and implementation. We turn performance audits into action-ready prompts so your team can move from report review to code changes with a lot less translation overhead.

Audit

See the issues that matter.

Real PSI scans, category scores, and finding-level evidence give you a trustworthy starting point.

Prompt

Package the work clearly.

Each finding is translated into language an engineer or coding agent can act on without guesswork.

Ship

Keep the feedback loop short.

Share result links, save scans in the dashboard, and rerun quickly after each round of fixes.

For Product Teams

Make perf work easier to prioritize.

Use saved scans, domain verification, and scan history to track where regressions are creeping in and where fixes are already paying off.

For AI-Assisted Dev

Give agents better starting context.

Instead of pasting a raw Lighthouse report into chat, start with a structured prompt that names the issue, the goal, and the likely code surface.

FAQ

Q.01

Is my scan private?

Anonymous scans are public by shareable link. Signed-in users can mark scans as unlisted so they stay out of search and sitemap feeds.

Q.02

Do I need an account to run a scan?

No. Public scans are available without sign-in. Accounts are for saved history, API keys, verified domains, and recurring workflows.

Q.03

What comes back in a result?

You get category scores, finding-level evidence, prompt copy, downloadable artifacts, and the same data the dashboard reads.

Q.04

Can I scan both Desktop and Mobile?

Yes. Every submission runs both strategies, and the results page lets you switch between them.

Q.05

Can my team use this outside the UI?

Yes. Signed-in users can create API keys and work with the same scan and artifact data programmatically.

Q.06

Can I keep checking a domain over time?

Yes. Verify a domain in the dashboard and use the recurring scan tooling to keep tabs on important pages over time.