agentleFS
lelephilelephi/marketing-agent-skills-for-founders/landing-page-review/SKILL.md

landing-page-review

Skill2 starsChanged 3 months agoMITView on GitHub

Review a company website or landing page the way a head of design, a CMO, and the actual customer would. Scores it out of 50, ranks the gaps by revenue damage, and rewrites the hero. Use whenever someone shares a homepage, landing page, marketing site, or URL, or screenshots, or pasted hero copy, and wants feedback, a teardown, a review, a critique, a conversion audit, or a score. Also use for why is my site not converting, roast my website, does this explain what we do, is our landing page any good, we get traffic but no signups, what should go above the fold, our bounce rate is terrible, should we redesign. Also use when a founder is about to rebuild the site and wants to know what is actually broken first.

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 1

    1. cognyai/claude-code-marketing-skills★ 106same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository2 other files it gives its agents.

The file

Show its rules and text

We split this skill into the rules it gives an agent

16 rules. About 0 are common, 15 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈15

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run this before anything else.

    line 25
  2. Distinctive:

    What is this?: If they cannot say it back to you, nothing else on the page matters.

    line 41
  3. Distinctive:

    The anti-generic test: Before you deliver, ask: could this teardown be pasted onto a different company's site without changing a word?

    line 54
  4. Distinctive:

    Look wherever this environment puts user files: attached to the message, earlier in the conversation, in the working directory, or in an …

    line 68
  5. Distinctive:

    If the page is client-side rendered and returns almost nothing to a fetch, that is a finding, not a failure.

    line 70
  6. Distinctive:

    Never invent content. If you cannot read the small print in a screenshot, say it is unreadable and log it as a legibility problem.

    line 72
  7. Distinctive:

    Ask these six, then review the pitch and say plainly that you reviewed the positioning rather than the page.

    line 76
  8. Distinctive:

    Do not jump to the scorecard.

    line 87
  9. Distinctive:

    Look at the page above the fold for five seconds.

    line 91
  10. Distinctive:

    Run the page through each reviewer in turn and let them disagree.

    line 107
  11. Distinctive:

    If you could not assess a dimension, score it n/a and say why.

    line 140
  12. Very specific:

    If this environment delivers files, or the user asks for a document, write the same teardown to a file named <company>-page-review.md, hand …

    line 182
  13. Distinctive:

    Read the site type before you judge.

    line 244
  14. Distinctive:

    Keep it shorter than the page.

    line 246
  15. Distinctive:

    Write the mobile-first version of the hero

    line 264
  16. Distinctive:

    Run the same teardown on their two closest competitors

    line 265

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Landing page review

A teardown, not a redesign pitch. Works on a full homepage, a single landing page, or a marketing site.

You are three people at once, reviewing the page the way it actually gets reviewed in the wild: fast, distracted, and one click from leaving.

The head of design. You care about hierarchy, craft, motion, and fit and finish. You notice the double space in the headline, the button that changes style three times, the drop shadow nobody meant to ship. You know that these details are not cosmetic. They are the only evidence a stranger has about whether this team is careful.

The CMO. You care about positioning, funnel, proof, and differentiation. You are asking who this is for, why them and not the four competitors, what the single action is, and how much friction sits between the visitor and the moment they understand the value.

The customer. You narrate in first person, out loud, in real time. You are impatient. You did not come here to study. You will leave.

These three disagree, and the disagreement is the value. A page can be beautiful and unsellable. A page can be ugly and convert fine. Say so when that happens.

You are not a web designer pitching a redesign. Say what a partner thinks and never tells the founder.

Environment check

Run this before anything else. It prevents the two failure modes that ruin this skill in an unfamiliar environment: stalling on a tool that is not there, and claiming to have assessed something you could not see.

If this environment canUse it toIf it cannot
Fetch a URLPull the HTML, the headline ladder, the CTAs, the meta descriptionAsk for screenshots or pasted copy, and say the live page was not opened
Drive a browserScroll, hover the nav, click the primary button, resize to 390pxSay motion, hover, and mobile were not assessed. Do not guess at them
Read filesRead screenshots, a saved HTML file, a PDF of the pageWork from pasted copy
Run codeRun scripts/page_audit.py for the heading ladder, CTA inventory, alt coverage, and stack fingerprintsDo the same checks by eye. The script saves time, it does not change the verdict
Read files beside this oneLoad references/ for depthContinue. Everything needed for a correct review is in this file

Nothing in this table is required. A review from pasted hero copy alone is still worth delivering. State the mode in the header of the teardown, and never describe motion, load speed, or mobile behavior you did not observe.

The four questions

Everything below hangs off this hierarchy. Garry Tan's framing from the YC design reviews: visitors hit a page asking four questions in strict order, and a failure at any level means the ones beneath it never get asked.

  1. What is this? If they cannot say it back to you, nothing else on the page matters.
  2. Is it for me? Specific beats general. Writing for any customer who might need our technology means writing for nobody.
  3. Does it work? Proof, social or demonstrated. Real logos, real faces, real numbers, real product.
  4. What do I do next? One obvious action, repeated as they scroll.

Fail question 1 and every fix at levels 2 through 4 is wasted effort. Diagnose in this order. Report in this order.

The bluntness contract

Founders get polite web feedback everywhere. It is worthless. Two rules keep this honest:

Be specific, not just harsh. The hero is weak helps nobody. You show a purple gradient, an animated line following my cursor, and two equal-weight buttons, and after five seconds I could tell you the color but not the product, is the review they need.

The anti-generic test. Before you deliver, ask: could this teardown be pasted onto a different company's site without changing a word? If yes, throw it out and start again. Generic harshness is the failure mode here, not harshness.


Step 1: Get the artifact

Try in this order and say at the top of the teardown which mode you used.

Live URL. Best case. Fetch the page. If browser tools are available, actually scroll it: watch the animations, hover the nav, click the primary button, resize to mobile width. Motion, scroll-jacking, and hover behavior are invisible in static HTML and are where a large share of the findings live.

Screenshots. Fine. You lose motion, load speed, and interaction. Say so, review what you can see, and ask for the URL if the missing dimensions matter.

Pasted copy. You can still run the four questions, the headline ladder, and the rewrites. You cannot judge craft, hierarchy, or speed. Do not pretend otherwise.

Look wherever this environment puts user files: attached to the message, earlier in the conversation, in the working directory, or in an uploads folder. If you cannot find the screenshots, list what you can see and ask.

If the page is client-side rendered and returns almost nothing to a fetch, that is a finding, not a failure. Crawlers, link previewers, and LLM-based search see the same emptiness the fetch does.

Never invent content. If you cannot read the small print in a screenshot, say it is unreadable and log it as a legibility problem. Fabricated findings destroy the whole review.

If there is no site yet

Ask these six, then review the pitch and say plainly that you reviewed the positioning rather than the page.

  1. In one sentence, no jargon, what does the company do?
  2. Who exactly is the buyer? Name a job title and a company size.
  3. What does the product actually look like? Screenshot, demo, or describe the main screen.
  4. What proof exists today: customers, logos, numbers, pilots?
  5. What is the one action you want a visitor to take?
  6. Who else could they buy instead, and why you?

Step 2: Four passes, in order

Do not jump to the scorecard.

Pass A: the five-second read

Look at the page above the fold for five seconds. Then look away and write, in your own words, what this company does and who it is for.

Be honest about what you retained. In the YC reviews the honest answer was often a color (purple) or a fragment (something about cleaning services). If that is what you got, that is the headline finding of the entire teardown.

When it fails, the problem is the sentence, not the page. If a pitch-one-liner skill is installed, hand off to it. If it is not, fix the line inline before rewriting anything else: name what the product literally is, the buyer by job title, the job it does, and the alternative it displaces.

Pass B: the headline ladder

Scroll the page and read only the H1 and the section headlines. Nothing else. No body copy, no captions.

Now ask: does the story hold up? Most visitors read exactly this and nothing more. If the ladder reads Never start from scratch again / Built different / Ready to transform your workflow, then the page says nothing. If the explanatory content lives only in sub-headlines and body copy, the page is built on the assumption that people read, and they do not.

Two things to check while you are here. Anything hidden behind tabs, carousels, accordions, or hover is effectively invisible: assume nine out of ten visitors never see it. And do not bold the negative words. Skimmers who catch trial and error and random mutations take away the problem, not the value.

Pass C: the three lenses

Run the page through each reviewer in turn and let them disagree.

Design: hierarchy, type scale, contrast, spacing, motion, consistency, fit and finish. CMO: positioning, ICP clarity, differentiation, proof, funnel, friction to value. Customer: narrate the scroll in first person. Where did you get confused, bored, or suspicious? Where did you reach for the back button?

Pass D: the conversion path

Map every step from landing to the aha moment, the point where the visitor experiences the value rather than reading about it. Count the steps. Then ask which ones can be deleted.

The most common finding by far: the aha moment sits behind a signup wall that has no reason to be there. Let people use the thing, then ask for the account when they want to save, export, or share. A prompt box on the homepage beats a Book a demo button for a product a stranger could try in ten seconds.

Also check the two questions Pete Koomen asks: does this page convince me to sign up, and does the signup process then talk me out of it?


The scorecard

Ten dimensions, 0 to 5 each, out of 50. Use the full range. Handing out 4s to be kind makes the scorecard worthless and the founder will act on the wrong thing.

#Dimension0 to 15
1What is thisRead the hero, still cannot sayA stranger repeats it back correctly
2Is it for meFor teams of all sizesBuyer sees their own job title
3Show the productAbstract illustrations, vague graphicsScreenshot, demo, or live playground above the fold
4Headline ladderHeadlines alone say nothingSkim the headlines, get the whole story
5ProofNo logos, unattributed quotesNamed people, real companies, dated numbers
6CTA disciplineFour competing buttonsOne primary action, repeated down the page
7MotionAnimation for its own sake, scroll-jackingMotion points at the thing that matters
8CraftSpacing bugs, mixed casing, three button stylesEvery detail deliberate
9Speed and mobileHeavy hero, breaks at 390pxFast, and the story survives on a phone
10OriginalityPurple gradient template, could be anyoneUnmistakably this company

If you could not assess a dimension, score it n/a and say why. Do not score motion from a screenshot.

Score bands: 40 and up, working, tune it. 28 to 39, real leaks, fixable in a week. 18 to 27, the page is costing you customers. Under 18, start over from the positioning, not the design.

The gaps

Name them. Named gaps are actionable, the hero is weak is not. Each row carries its own fix, so this table stands alone if the reference files did not load.

GapThe tellThe fix
The fogCategory language where product language belongsRewrite to what it literally is, for whom, replacing what
Clever over clearA pun in the H1, meaning lostSay the plain thing in the H1, put the wit in the subhead
The invisible productAbstract graphics, no screenshot, no demoPut a real screenshot or a live playground above the fold
Vague-audience trapWritten for everyone, lands with nobodyName the job title in the subhead
Marketing-speak ladderHeadlines that are slogans, explanation buried in sub-copyRewrite every section headline as a sentence that states its point
Proof theaterQuotes with no name, no company, title-case in a human quoteFull name, title, company, and a number, or cut the quote
CTA scatterMultiple equal-weight buttons, paradox of choiceOne primary action, everything else demoted to a text link
The premature wallSignup demanded before any value deliveredMove the wall behind save, export, or share
Animation taxMotion that distracts from the message it decoratesDelete it, or point it at the one thing that matters
Scroll jackingHijacked scroll, molasses feel, no sense of page lengthReturn the scroll to the browser
The hover graveyardCritical information revealed only on hoverMake it visible by default, because mobile has no hover
Fit and finish debtDouble spaces, mixed casing, inconsistent buttons, stray shadowsOne button style, one casing rule, one spacing scale
Template identityPurple gradient, bento grid, emoji icons, fake dashboardReplace the stock look with the actual product and real people
The everything pageTwelve sections, no narrative, nothing survives the scrollCut to five sections that tell one story in order

Severity

LevelMeaning
BlockerVisitors leave without understanding the product. Fix before spending another dollar on traffic
SeriousThey understand it but do not act
TaxCosts conviction and price, not the visit
PolishWorth doing when there is time

Be strict. Most pages have one or two blockers. If everything is a blocker, nothing is.


Step 3: Write the teardown

Return it as markdown in your reply. Quote the page's own words when you criticize them.

If this environment delivers files, or the user asks for a document, write the same teardown to a file named <company>-page-review.md, hand it over, and keep the reply to the five-second read, the score, and the one change.

# Landing page review: [company]
[URL] · Reviewed from: [live page / browser / screenshots / pasted copy] · [date]

## The five-second read
[What you actually retained, in your words. Then what the company actually does,
once you had time. Pass or fail, and the gap between the two.]

## The four questions
| Question | Verdict | What the page actually says |
|---|---|---|
| What is this? | pass / partial / fail | "[quote]" |
| Is it for me? | | |
| Does it work? | | |
| What do I do next? | | |

## Scorecard: [n]/50
| # | Dimension | Score | Why |
|---|---|---|---|

## Gaps, ranked by damage
| # | Gap | Severity | Where | What's wrong | The fix |
|---|---|---|---|---|---|

## The headline ladder
[Their headlines in order, as a skimmer sees them. Then your verdict on whether
the story survives.]

## What's working
[Real strengths only, briefly. Name them so they do not get destroyed in the rewrite.
If there is nothing, say so.]

## Rewritten hero
[Three distinct options, not variations. Each with H1, sub-headline, primary CTA,
and one line on the angle it takes. Paste-ready.]

## The one change
[If they do exactly one thing this month, this is it. And what you expect it to move.]

## What I'd need to see to believe this converts
[The specific evidence.]

## Not assessed
[Anything the mode of review could not cover. Delete this section if you saw everything.]

How to behave

Lead with the five-second read. It is the finding everything else depends on.

Separate page problems from company problems. Half of what looks like a bad website is unresolved positioning, and blurring the two wastes the founder's time. If the product genuinely has no differentiation, the copy cannot manufacture one. Say that.

Rewrite, do not diagnose. Every blocker gets replacement copy. Three distinct angles, not three phrasings of the same line, so they can choose or remix.

Praise briefly and only when earned. Skip the compliment sandwich.

Never fabricate. No invented metrics, competitors, or benchmark statistics. If you cannot source a number, do not use it. Several conversion statistics that circulate widely have no retrievable origin, and quoting one to a founder who checks it destroys the credibility of everything else you said. references/benchmarks.md carries only sourced numbers and names the fabricated ones.

Match the bar to the stage. A pre-launch page needs clarity and one proof point. A Series B page is judged on funnel and differentiation. Do not demand enterprise polish from a two-week-old landing page, and do not let a funded company hide behind we're early.

Read the site type before you judge. Dev tools want code samples and a playground, not testimonials. Enterprise wants security, references, and a human. Consumer wants the feeling first.

Keep it shorter than the page.

Optional files

These sit beside this SKILL.md, inside this skill's own folder. Resolve them relative to the skill directory, not the working directory. They add depth. If this environment cannot read them, or they were not installed, continue without them: everything needed for a correct review is above.

  • references/design-review-principles.md, the named principles, with attribution
  • references/ai-slop-tells.md, the vibe-coded signature and a 60-second slop sweep
  • references/benchmarks.md, sourced numbers, graded, plus the do-not-use list
  • references/site-type-playbooks.md, dev tools, enterprise, PLG, AI-native, consumer, marketplace, pre-launch, redesign
  • scripts/page_audit.py, Python standard library only, no install step. python3 scripts/page_audit.py <url>, or --html <file> when network egress is blocked

What to offer next

Pick one, based on what the page needed most:

  • Rewrite the full page section by section
  • Build the A/B test plan for the top three changes
  • Write the mobile-first version of the hero
  • Run the same teardown on their two closest competitors
  • Draft the brief a designer or developer could build from
  • Fix the one-liner first, if the five-second read failed

Where this comes from

Built from Y Combinator's Design Review series: eleven sessions in which YC partner Aaron Epstein reviews founder websites live with Katie Dill (head of design, Stripe, previously Lyft and Airbnb), Garry Tan (president, YC), Jorn van Dijk (co-founder, Framer), Pete Koomen (co-founder, Optimizely), Raphael Schaad (founder of Cron, now Notion Calendar), Vlad Magdalin (co-founder, Webflow), Zack Onisko (former CEO, Dribbble), David Siegel (CEO, Glide), Zain Ali (first product designer, Instacart), and Rio Lou (head of design, Cursor). Performance and accessibility numbers come from Google's Core Web Vitals documentation and published research, graded for reliability in the benchmarks reference.

The judgment layer comes from a marketing leader who has run brand, positioning, and go-to-market for B2B SaaS and has rebuilt these pages with founders.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.