Run any URL through Google PageSpeed Insights, then download a CLAUDE.md tailored to the failing audits. Drop it into your repo and let Claude Code apply the fixes.
The annoying moment that started this: I was asking Claude Code to "make this page faster," then pasting a PageSpeed Insights screenshot, then spending ten minutes spelling out which files to look at, what each cryptic audit ID actually meant ("max-potential-fid" really?), and which audits not to break. Every. Single. Time.
It felt like I was building the prompt manually when the data was already structured. So I built the missing middle layer: paste a URL, get a CLAUDE.md that Claude Code can act on directly.
What it actually contains (this is the part I'm most proud of):
- The throttling profile Lighthouse used (so Claude doesn't over-optimize for desktop when you measured on mobile) - Every failing audit with a concrete fix suggestion, not just the metric - The actual offending URLs, selectors, and HTML snippets, ranked by potential savings - Third-party impact table, main-thread time breakdown, JS bootup time per script - A "DO NOT BREAK" list of every currently-passing audit - Stack-pack guidance (Lighthouse knows you're on WordPress / React / Next.js and gives stack-specific tips)
How the approach evolved: v1 was just "scores + failing audit list." Useless, Claude didn't know what files to look at. v2 added the offender lists. Better, but Claude would sometimes regress passing audits to fix failing ones. v3 (this launch) added the test environment, passing-audits constraint list, and stack-pack guidance. That's when the output started feeling like a brief instead of a report.
There's also a one-click "Ask Claude to fix problems" button that opens Claude Code via the claude://code/new deep link with the entire brief pre-loaded, that's the workflow I personally use.
Free, no signup, no API key required from users.
What I'd love to know from you:
Which Lighthouse audits do you actually fix vs. ignore? I want to weight the suggestions better. Would historical tracking (analyze the same URL weekly, see deltas) be useful, or is one-shot enough? Happy launch day, and thanks for taking a look 🙏
Community Inviter
Hey Product Hunt! Maker here 👋
The annoying moment that started this: I was asking Claude Code to "make this page faster," then pasting a PageSpeed Insights screenshot, then spending ten minutes spelling out which files to look at, what each cryptic audit ID actually meant ("max-potential-fid" really?), and which audits not to break. Every. Single. Time.
It felt like I was building the prompt manually when the data was already structured. So I built the missing middle layer: paste a URL, get a CLAUDE.md that Claude Code can act on directly.
What it actually contains (this is the part I'm most proud of):
- The throttling profile Lighthouse used (so Claude doesn't over-optimize for desktop when you measured on mobile)
- Every failing audit with a concrete fix suggestion, not just the metric
- The actual offending URLs, selectors, and HTML snippets, ranked by potential savings
- Third-party impact table, main-thread time breakdown, JS bootup time per script
- A "DO NOT BREAK" list of every currently-passing audit
- Stack-pack guidance (Lighthouse knows you're on WordPress / React / Next.js and gives stack-specific tips)
How the approach evolved: v1 was just "scores + failing audit list." Useless, Claude didn't know what files to look at. v2 added the offender lists. Better, but Claude would sometimes regress passing audits to fix failing ones. v3 (this launch) added the test environment, passing-audits constraint list, and stack-pack guidance. That's when the output started feeling like a brief instead of a report.
There's also a one-click "Ask Claude to fix problems" button that opens Claude Code via the claude://code/new deep link with the entire brief pre-loaded, that's the workflow I personally use.
Free, no signup, no API key required from users.
What I'd love to know from you:
Which Lighthouse audits do you actually fix vs. ignore? I want to weight the suggestions better.
Would historical tracking (analyze the same URL weekly, see deltas) be useful, or is one-shot enough?
Happy launch day, and thanks for taking a look 🙏