agentleFS
mardab96mardab96/landing-pages-claude-skills/AGENTS.md

landing-pages-claude-skills

AGENTS.md2 starsChanged 4 months agoMITView on GitHub

Guidance for the agent running these skills, and for anyone extending the pack. A landing page diagnosis is cheap to produce and expensive to be wrong about, because it gets shipped. So the pack is built so that a skill running on thin input says the input is thin instead of producing a confident answer that sounds identical to a well-founded one. If you take one behaviour from this file: name the assumption in the output, at the point where…

The file

We split this AGENTS.md into the rules it gives an agent

7 rules. About 0 are common, 5 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If you take one behaviour from this file: name the assumption in the output, at the point where it affects the recommendation.

    line 12
  2. Very specific:

    Ends on a decision: Every skill closes with Decision: and a short list of named options, so the output ends somewhere a human can act.

    line 22
  3. Distinctive:

    Recommends the smallest change that removes the defect: Not a redesign.

    line 36
  4. Distinctive:

    Never executes: No page edits, no campaign changes, no CRM writes.

    line 37
  5. Distinctive:

    Run them. Quote what they print.

    line 97
  6. Distinctive:

    Keep the section order: Use this skill when, Required input, Analysis workflow, Decision rules, Output format, Practical example, …

    line 105
  7. Very specific:

    Run scripts/validate-skills.py and scripts/selftest.py before publishing.

    line 113

≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md 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
AGENTS.md

How this pack behaves

Guidance for the agent running these skills, and for anyone extending the pack.

The one rule that matters

A landing page diagnosis is cheap to produce and expensive to be wrong about, because it gets shipped. So the pack is built so that a skill running on thin input says the input is thin instead of producing a confident answer that sounds identical to a well-founded one.

If you take one behaviour from this file: name the assumption in the output, at the point where it affects the recommendation. Not in a disclaimer at the end.

What every skill does the same way

  • Reads the evidence before interpreting it. State what the data shows, then what it means. A reader must be able to disagree with the interpretation while accepting the evidence.
  • Marks missing input where it bites. "Scroll data was not supplied, so the ranking below is judgement" beats a footnote nobody reads.
  • Ends on a decision. Every skill closes with Decision: and a short list of named options, so the output ends somewhere a human can act.
  • Labels confidence. High, medium, low, unknown. Unknown is a legitimate and frequently correct answer.
  • Labels heuristics as heuristics, and says which kind. Some carry a real source and can be cited: 4.5:1 and 3:1 contrast are WCAG 2.1 AA, 44 by 44 points is Apple's guidance, 48 by 48 dp is Google's. Others are this pack's own working defaults with no external source at all: the 35 word long-sentence flag, the 20 word mean, the ten-sentence floor below which a mean is not adjudicated, the 150 word floor for the frequency table, the three-decisions-per-screen ceiling, the 25th percentile scroll cutoff, the 100 conversions per arm refusal gate, and the seven day runtime gate. Say which of the two a number is every time you quote one, and never imply a source that does not exist. Where a real measurement exists, it beats both kinds.
  • Recommends the smallest change that removes the defect. Not a redesign.
  • Never executes. No page edits, no campaign changes, no CRM writes.

Composition: which skills run together

The pack is deliberately overlapping at the edges, because the same symptom has several possible causes. Useful chains:

  • Cold traffic that does not convert: paid-traffic-message-match-audit sets whether the promise survived the click, offer-clarity-diagnosis whether the promise is worth anything, traffic-temperature-match-review whether the page is even being asked to serve one audience. Run in that order; stop when one of them produces a blocker, because the later findings will be about the wrong page.
  • Form problems: form-friction-finder for the mechanics of stopping, lead-form-sales-handoff-check for whether the fields deserve to exist, trust-signal-audit for the hesitation right before submit. All three read the same form and answer different questions.
  • Mobile gap: mobile-conversion-review, then popup-and-overlay-timing-review, then accessibility-conversion-blocker-check. Overlays and tap targets are the two most common real causes behind a mobile-versus-desktop gap.
  • Before scaling spend: landing-page-scale-readiness-check and traffic-temperature-match-review together. The first asks whether the page holds, the second whether the current rate transfers to the new traffic.
  • Before publishing anything public-facing: comparison-page-positioning-review for claims, accessibility-conversion-blocker-check for defects, both of which can stop a ship.

Where the boundaries are

Overlap is intentional; duplication is not. If a new skill would answer a question one of these already answers, extend that skill instead.

If the question isIt belongs to
Does the ad match the pagepaid-traffic-message-match-audit
Does the page match the audience temperaturetraffic-temperature-match-review
Why do people abandon the formform-friction-finder
Should this field existlead-form-sales-handoff-check
Is the copy hard to readlanding-page-copy-readability-pass
Is the offer vagueoffer-clarity-diagnosis
How long should the page bepage-length-fit-check
Which ask should the page maketrial-vs-demo-path-decision
Is the CTA wording clearcta-clarity-check

Tracking, pixels, GA4 and conversion plumbing are out of scope for this pack. They are a different problem with different failure modes.

Three helpers, and why they exist

significance.py, readability_report.py and contrast_check.py exist because a language model estimating a p-value, a contrast ratio or a mean sentence length will produce a plausible number, and a plausible number is indistinguishable from a measured one in the output. significance.py matters most of the three: it is the only place in the pack where a threshold refuses a conclusion the operator wants, and it enforces that in an exit code rather than in prose. All three live inside the skill that cites them, so an installed skill never points at a path the install command forgot to copy.

Run them. Quote what they print. Do not restate their numbers from memory.

Extending the pack

  • Frontmatter needs a name and a description whose second half names the situation that should trigger the skill. Do not reuse another skill's activation sentence. The whole pack shared one once, and it made routing between 15 skills a coin flip.
  • Keep the section order: Use this skill when, Required input, Analysis workflow, Decision rules, Output format, Practical example, Guardrails.
  • The workflow needs at least five steps, and each step must be an action on the input rather than a topic to consider.
  • The practical example carries real numbers and a real verdict, including what the skill refuses to conclude. A generic example teaches the model to produce generic output.
  • Anything a script can count, a script should count.
  • Run scripts/validate-skills.py and scripts/selftest.py before publishing.

references/output-standard.md holds the shared output bar and the evidence tags. references/skill-design-principles.md holds the design rules. references/industry-notes.md holds the places where vertical changes the judgement.

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.