iiinova — where your digital self keeps evolving

I don’t turn single outliers into tests. Tests come from segments with volume: p75 device, connection and the top paths by traffic. Outliers go to investigation first; if the pattern repeats across sessions, then it earns a test.

100

Joan León @nucliweb.net · 21h

100%. The goal is lab tests that look like real users: same devices, same conditions, same paths. The MCP is making that loop really fast.

100

Reposted by Joan León

Ben Schwarz @benschwarz.bsky.social · 29/09/2026

How far ahead browsers fetch lazy images: • Chrome: 1,250px on 4G, 2,500px on 3G, 6,000px (!!) on 2G • Firefox: 600px, on any connection • Safari: one viewport. Lazy iframes: 0px, they wait until on screen

calibreapp.com

Lazy Loading Images and Media: How Far Ahead Browsers Fetch

How far ahead Chrome, Firefox and Safari fetch lazy media, why a lazy hero image slows your page, and where to use lazy, eager and fetchpriority.

0118

Joan León @nucliweb.net · 29/09/2026

One thing I’m experimenting with for a performance audit: Using RUM data + @debugbear.com MCP to build synthetic tests based on how the product is actually used RUM → real user behaviour → MCP → representative synthetic tests www.debugbear.com/docs/mcp #WebPerf #RUM #MCP

debugbear.com

Access Web Performance Data With The DebugBear MCP Server | DebugBear

Connect AI assistants like Claude and ChatGPT to DebugBear using the Model Context Protocol (MCP) to run tests, query metrics, and manage projects.

330

Joan León @nucliweb.net · 20/09/2026

The case against JPEG XL Investigating JPEG XL's place as a Web image codec, by @giannirosato.com, via @webperfreport.bsky.social giannirosato.com/blog/post/ca... #JPEGXL #JXL #ImagePerf #WebPerf

A dark-themed line chart titled “SSIMULACRA2,” showing BD-Rate (%) versus encoding speed in milliseconds for several image codecs. Colored lines compare codecs including AOM, JPEG XL, aperture-alpha, Iris-WebP, libjxl, SVT-AV1, and libwebp, with labeled data points. Most curves show higher BD-Rate at slower speeds, while some reach negative BD-Rate at faster speeds.

032

Joan León @nucliweb.net · 16/09/2026

I added WebMCP to the GeeksCAT Conf 2026 website, so an agent can ask for the schedule, the speakers or the event details as tools instead of reading the HTML. It's a static Astro site, so I go through it layer by layer. joanleon.dev/en/webmcp-st... #WebMCP

joanleon.dev

WebMCP on a static site: the GeeksCAT Conf 2026 website, ready for agents | Joan León

The GeeksCAT Conf 2026 website is a static Astro site that exposes its schedule, speakers and event details to AI agents through WebMCP. Three layers: a manifest in /.well-known, JSON endpoints genera...

153

Reposted by Joan León

Chrome for Developers @developer.chrome.com · 16/09/2026

We created Modern Web Guidance to bring web platform expertise right into your agent workflows. Have you tried pairing these skills with Chrome DevTools for agents to audit performance and automatically fix code? Check it out and join 10K+ web developers who are coding smarter → goo.gle/4yEVz1h

181

Joan León @nucliweb.net · 07/09/2026

P75 is green: LCP 1.58s, INP 176ms, CLS 0.01. P90: LCP 2.68s, INP 352ms, CLS 0.17. 15% of visits get a noticeably worse experience. Product Performance doesn't end once you pass CrUX: if P75 is good, work on P90. #WebPerf #CoreWebVitals #RUM #ProductPerformance

DebugBear RUM dashboard showing Core Web Vitals at the P75 percentile: Largest Contentful Paint 1.58s, Interaction to Next Paint 176ms, and Cumulative Layout Shift 0.01, all flagged green on the score bar.

DebugBear RUM dashboard showing Core Web Vitals at the P90 percentile: Largest Contentful Paint 2.68s, Interaction to Next Paint 352ms, and Cumulative Layout Shift 0.17, all flagged orange or red on the score bar.

073

Reposted by Joan León

WebPerformance Report @webperfreport.bsky.social · 06/09/2026

⚡Featured posts this week: ✨ Submit your proposals for Interop 2027: @rachelandrew.co.uk ✨ Reading a performance profile: @nucliweb.net ✨ HTML over WebSockets: @androsfenollosa.bsky.social ✨ Intro to Web UX and Speed Optimization: @sergeyche.dev 🙏 Special thanks to the authors! #webperf

WebPerformance Report performance email preview highlighting optimization scores, Perf Links, and Core Web Vitals metrics on a colorful gradient background.

062

Reposted by Joan León

Thomas Steiner @tomayac.com · 28/08/2026

🤩 Fantastic article by @nucliweb.net on HTTP cache partitioning and the #CrossOriginStorage API: joanleon.dev/en/http-cach.... Try the embedded demos with the COS extension installed or in one of the experimental browser builds (see the Awesome Cross-Origin Storage list: github.com/tomayac/awes...).

joanleon.dev

HTTP cache partitioning: the same image, downloaded twice | Joan León

A page with a cross-site iframe requests the same image from the iframe and from the parent. Chrome downloads it twice; Firefox and Safari, once. Why the cache key differs across the three engines, wh...

0153

Reposted by Joan León

WebPerformance Report @webperfreport.bsky.social · 16/08/2026

⚡Featured posts this week: ✨ AI and organic search are doing different jobs: @kylerisley.bsky.social ✨ The CrUX 28-day rolling window: @nucliweb.net ✨ AI Web Readiness as a Curb-Cut: @ethangardner.com ✨ HTTP Integrity Digests: @simmervig.org 🙏 Special thanks to the authors! #webperf

031

Joan León @nucliweb.net · 16/08/2026

I love the optimization: web, rendering, images, video, and now tokens. @lydiahallie.com published a great guide to Maximizing the value of your Claude Code sessions claude.com/blog/maximiz... #AIOptimization #AI #Claude

A dark-themed presentation slide titled "TL;DR" displaying tips for optimizing Claude API usage and token efficiency. The slide is divided into two columns:

Left column discusses:
- Using `/clear` command between tasks to prevent context carryover and reduce token usage
- Setting model and effort level before starting a session to avoid breaking the prompt cache mid-conversation
- Mentioning files with @ notation instead of naming them, to attach files directly and save API calls

Right column covers:
- Adding quiet flags to commands or running them in a subagenr to manage noisy output while keeping results in the conversation
- Running `/context` once per session to review loaded content (CLAUDE.md and MCP tool definitions) and remove unnecessary items
- Running `/compact` before breaks, since prompt cache expires after an hour and summarizing conversations is cheaper while the cache is active

070

Reposted by Joan León

Martin Michalek @machal.bsky.social · 14/08/2026

Reading a Performance Timeline by @nucliweb.net. This would be great for people who still have issues with Performance Timeline in Chrome. I love the form of this site. perf.reviews/profile-guide

perf.reviews

Reading a Performance Profile: Field Guide

An interactive field guide to the Chrome DevTools Performance panel: waterfall, flame chart, LCP, CLS, INP, and the Insights panel.

0235

Joan León @nucliweb.net · 13/08/2026

Most skills are a Markdown file that grows out of control and gets loaded whole, even when the task doesn't need it. Modern Web Guidance inverts the pattern: 250 fixed tokens, and the agent retrieves the guide it needs on demand. joanleon.dev/en/modern-we... #AI #WebDev #CodingAgents

joanleon.dev

A skill that knows nothing: the architecture of Modern Web Guidance | Joan León

Modern Web Guidance doesn't keep the knowledge in the skill, it teaches the agent to retrieve it. Local semantic search with MiniLM and TensorFlow.js, guides with their own grader, and empirical calib...

061

Joan León @nucliweb.net · 11/08/2026

In my audits, the question I hear most after a deploy is "why hasn't CrUX moved yet?" The window is always 28 days; daily, weekly and monthly are just cadences over the same data. And p75 resists change by design I explain it with two interactive demos joanleon.dev/en/crux-28-d... #WebPerf #CrUX

joanleon.dev

The CrUX 28-day rolling window | Joan León

How the CrUX 28-day rolling window actually works, what each source publishes, and why p75 takes so long to move even though the data is only two days old.

0104

Joan León @nucliweb.net · 05/08/2026

A letter from @addyosmani.bsky.social arrived today, thanking me for my work on Agent Skills. I got in through web performance and stayed for the rest. Thinking about contributing to open source? Start small. Docs and honest reviews count. github.com/addyosmani/a...

A printed thank-you letter from Addy Osmani on a dark background, hand-signed and dated 2026, next to its opened envelope postmarked San Francisco. Four stickers lie beside it: "Coder", "Computer Club", a Cmd+Z one reading "Just Undo It", and a perfume bottle reading "#perfmatters".

000

Joan León @nucliweb.net · 04/08/2026

Isolated spikes off a stable ~2s baseline, different days, different countries. Not one user on a bad connection, and not bots, I checked. Which is the whole point of RUM: synthetic checks on a schedule would have missed every one.

SpeedCurve budget dashboard: status "Over", latest value 27.93s against a 0.15s budget, 18,520% over. The weekly chart shows a stable baseline around 2s with sharp isolated spikes above 20s on Jul 29, Jul 31, Aug 3 and Aug 4.

000

Joan León @nucliweb.net · 04/08/2026

The site is static, on Netlify. 27s to first byte means the request never resolved at the edge: cache miss to origin, cold start, or a failing PoP.

SpeedCurve size metrics: HTML 15 KB, inline styles 14 KB, inline JS 8 KB. Below, the sessions table lists a session at 7:06 AM from the United States on Chrome 149, with an average First Contentful Paint of 27.93s.

100

Joan León @nucliweb.net · 04/08/2026

Update: this isn't frontend. TTFB is 27.44s of a 27.93s FCP. 98% of the time is spent waiting for the first byte, and the HTML is 15 KB.

SpeedCurve time metrics: Backend (TTFB) 27.44s, Frontend 8.18s, Start Render 27.93s, First Contentful Paint 27.93s, Largest Contentful Paint 27.97s, Page Load 35.62s. First Input Delay and INP both 0 ms.

100

Joan León @nucliweb.net · 03/08/2026

A fast site isn't a project you close on launch day. It's a state you have to maintain. The alert is the beginning, not the end. Now: root cause, and give users back the experience they expect.

111

Joan León @nucliweb.net · 03/08/2026

@speedcurve.com alert, fired on its own: FCP (RUM) budget breached, metric at 4.72s. Flat and green for hours, then climbing. Nobody was watching a dashboard. Synthetic gives reproducibility. RUM gives the truth: real users, real devices, real networks.

SpeedCurve alert: the First Contentful Paint (RUM) budget exceeds its 0.15s threshold by 3047%, reaching 4.72s. The chart shows the metric flat and green until 20:00, then climbing and spiking to nearly 5s.

110

Joan León @nucliweb.net · 03/08/2026

Web performance is a marathon, not a sprint. Ship the fastest site today and in two weeks a third-party script, an unoptimised image or a CDN change will have quietly undone it. Optimising without monitoring doesn't last. 🧵 #WebPerf #RUM #CoreWebVitals

181

Reposted by Joan León

Andrey Sitnik @en.sitnik.es · 30/07/2026

While some people laughed about the number of dependencies in the JS world, others worked to improve the ecosystem (like @e18e.dev). Now many of my open source projects have around 20 packages in node_modules, including transitive dependencies.

5577

Joan León @nucliweb.net · 26/07/2026

Fair callout, and thanks for it. You're right, so I went back and did that second pass: cut the aphorism-y openers (those two included) and reworked the section intros to read straighter, in a voice closer to how I actually talk through a trace. It's live on the guide now. Thanks for the feedback!

010

Reposted by Joan León

WebPerformance Report @webperfreport.bsky.social · 26/07/2026

⚡Featured posts this week: ✨ Is your site agent-ready?: @cloudflare.social ✨ Measuring soft navigations: @tunetheweb.com and @mmocny.com ✨ Reading a performance profile: @nucliweb.net ✨ The Absolute State of Management: @infrequently.org 🙏 Special thanks to the authors! #webperf

231

Joan León @nucliweb.net · 25/07/2026

Thank you Jack ☺️

000

Reposted by Joan León

James @43081j.com · 19/07/2026

We got a few new individual sponsors recently in @e18e.dev - thank you to those of you who are supporting us 🙏 Getting sponsorships for a community is much harder than for a popular web app or dev tool. Every little helps. If you'd like to help us improve perf for all of us, check out our OC 🩵

opencollective.com

e18e - Ecosystem Performance - Open Collective

The e18e community has the aim of improving performance of tooling and packages in the JavaScript ecosystem.

13811

Joan León @nucliweb.net · 24/07/2026

In my audits, reading the Performance profile is where people get stuck: the flame chart and waterfall look like noise at first I built an interactive guide to make them click, every figure is interactive, plus LCP, CLS, INP and the Insights panel perf.reviews/profile-guide #WebPerf #DevTools

perf.reviews

Reading a Performance Profile: Field Guide

An interactive field guide to the Chrome DevTools Performance panel: waterfall, flame chart, LCP, CLS, INP, and the Insights panel.

38223

Joan León @nucliweb.net · 19/07/2026

Thank you Jack ☺️ I’ll think about it… I’ve some ideas 😉

110

Reposted by Joan León

Sergey Chernyshev @sergeyche.dev · 19/07/2026

The curse of web performance - once you know it, you can’t unknown it and I see performance issues on the sites I visit too and can’t help it but open DevTools often as well. I was even thinking for creating a podcast-style Meet4SPEED YouTube channel looking at websites and improving them.

151

Joan León @nucliweb.net · 19/07/2026

This is how PerfReviews began, doing site reviews, with the aim of sharing how to analyse and detect improvement options

020

Joan León @nucliweb.net · 19/07/2026

An occupational habit: I open DevTools on every site I visit, hunting for things to improve. On skills.addy.ie I caught a 301 on every internal click, a wasted round trip. It was a trailing-slash mismatch. Wrote up the finding and the fix: joanleon.dev/en/trailing-... #WebPerf #Astro #Netlify

Screenshot of a desktop web browser displaying the agent-skills website on the left and Chrome DevTools’ Network tab on the right. The webpage heading reads “How agent-skills compares” with explanatory text describing how agent-skills relates to Superpowers and Matt Pocock’s skills, along with a “TL;DR” summary card. The DevTools Network panel shows document requests, including several 301 Moved Permanently redirects (such as skills, getting-started, lifecycle, teach, and compare) followed by corresponding 200 OK responses for URLs ending with trailing slashes. The browser address bar displays skills.addy.ie/compare/, and the Network panel is configured with Preserve log and Disable cache enabled and No throttling selected.

261

Joan León @nucliweb.net · 17/07/2026

Simple test: every conclusion needs an `evidence` field pointing to real data (Lighthouse, RUM, waterfall), not generated text. If the agent can't cite it, it doesn't make the report. Testing it with any specific workflow?

110

Joan León @nucliweb.net · 17/07/2026

And if you have the opportunity to run it on the source code, the potential is impressive. I'll write a post with my experience and thoughts.

100

Joan León @nucliweb.net · 17/07/2026

I'm testing the @debugbear.com MCP with Claude Code, and my first impression is that combining DebugBear's great tools with the power of AI Agents is awesome for creating executive and detailed technical reports... #WebPerf #PerfTools #AI #WebPerfAI

210

Joan León @nucliweb.net · 14/06/2026

I've been running #Antigravity Skills in real projects and the YAML frontmatter does much more than the docs show. kind: subagent + restricted tools + the right model per task. That trio makes agent behavior predictable and keeps costs in check. joanleon.dev/en/antigravi... #AI #AgentSkills

joanleon.dev

Antigravity Skills: advanced YAML frontmatter configuration | Joan León

The kind, model, temperature, max_turns and tools fields in SKILL.md frontmatter let you precisely control how each agent runs in Antigravity. A practical guide to the extended fields that Google hasn...

010

Joan León @nucliweb.net · 07/06/2026

Friend: Hey, Joan, can you give us a hand? Me: Sure, send me the URL DevTools: It takes nearly 3 seconds to load the WordPress plugins Me: 🫠

1151

Joan León @nucliweb.net · 07/06/2026

Este fin de semana he podido conocer a Edu, ver su Keynote en @commit-conf.com, y conversar con él, y tengo muy claro que quiero leer ese libro 📖

041

Joan León @nucliweb.net · 06/06/2026

Gracias @commit-conf.com por darme la oportunidad de compartir con la comunidad Estas son las slides de la charla slides.com/joanleon/pat... #CommitConf #WebPerf

050

Reposted by Joan León

WebPerformance Report @webperfreport.bsky.social · 31/05/2026

⚡Featured posts this week: ✨ Zero Trust Implementation Guidelines: National Security Agency ✨The Website Specification: @joost.blog ✨Agent Skills: @addyosmani.bsky.social @oreilly.bsky.social ✨Lighthouse agentic browsing scoring: @developer.chrome.com 🙏 Special thanks to the authors! #webperf

121

Reposted by Joan León

Chrome for Developers @developer.chrome.com · 28/05/2026

AI coding tools are powerful at writing code, but they can't inspect execution in a live browser. The stable 1.0 release of Chrome DevTools for agents provides your AI assistant with the visibility it needs to verify, debug, and optimize code in real time → goo.gle/42K7Rrl #GoogleIO

0123

Joan León @nucliweb.net · 28/05/2026

“Most teams fall into a reactive performance decay cycle: ship, wait for complaints, panic, patch, repeat. The Fast by Default model provides a repeatable alternative that keeps systems healthy.” Performance Engineering in Practice, by @denodell.bsky.social #WebPerf #Performance

Slide titled “1.6 Summary” from a presentation called “Performance Engineering in Practice.” The slide contains bullet points about software performance engineering. A highlighted bullet states that many teams fall into a reactive performance decay cycle: they ship software, wait for complaints, panic, patch issues, and repeat the cycle. It adds that the “Fast by Default” model offers a repeatable alternative that keeps systems healthy. The slide has a light gray background with dark blue and black text, and the page number “74” appears at the bottom center.

361