Index of every page on this site: llms.txt
All skills
Run it yourself

Rebuilds a named reference from its real source and computed values, then compares the two side by side.

Use this Skill: https://skilld.dev/gh/raphaelsalaja/skills/match

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈28 tokens for metadata: the name and description. ≈1.6k when used: this file. ≈24 more on demand in 1 file.

Description uses 1.4% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈112 characters, or 1.4%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Match

This skill rebuilds something the user pointed at in another app, site or design. It works from the real source and its measured values, not from memory of how it looks.

One to one means exact values

When the user says "one to one", "exact" or "like this", every value comes from the reference. A guessed value is a finding. When the user says "inspired by", carry the concept and use the project's own tokens.

Use it whenever a message names an app, a URL, a screenshot, a design file or a design reference library. Building with the project's parts belongs to reuse. Breaking a reference's motion into tracks on a timeline belongs to deconstruct. Checking the result runs in verify.

1. Open the real source

Open the reference itself: the live page, the installed app, the design file or the screenshot. Rebuilding from memory is match-inspect-the-source. When the reference is a native or desktop app, look at that app, not its website. Judging a desktop app by its marketing page is match-same-platform.

2. Read the values

Read computed styles, design tokens or measurements from the source. Write down sizes, spacing, radius, colours, type, durations and easing. A radius chosen by eye is match-read-computed-values.

Read each value once motion settles, when el.getAnimations({ subtree: true }) returns an empty list. A computed value mid-transition is one frame, not the value. Read getComputedStyle(el, "::before"), "::after" and "::placeholder" too, since borders, icons and fades often live there.

Read the font each side actually renders with the DevTools protocol's CSS.getPlatformFontsForNode, not the first name in font-family. A rebuild that copies the stack but renders Arial, because the first font never loads in the project, is match-rendered-font.

Read motion from the running animations. Trigger the change, then read getTiming() and getKeyframes() from each entry in el.getAnimations({ subtree: true }). A guessed 200ms ease on a menu that runs 240ms cubic-bezier(0.32, 0.72, 0, 1) is match-read-the-motion.

A design file's type values need converting. Letter spacing of -2% is -0.02em, and a line height of 150% is 1.5. letter-spacing: -2% pasted into CSS is dropped by the browser, and is match-convert-design-units.

A screenshot measures in device pixels. Find its scale from the image width against the window, such as 2880px for a 1440 window, then divide every measurement by it. A 24px gap read off a 2× screenshot copied as gap: 24px is match-measure-at-scale. Sample a colour from a flat patch at least 4×4 pixels away from any edge, and mark it Not Verified.

3. Inventory the states

List every state the reference has: hover, pressed, focus, open, empty, loading, error, selected and disabled. Then list its transitions. Rebuilding the resting state alone is match-inventory-states.

Hold hover, focus and active with the DevTools protocol's CSS.forcePseudoState, then read the values again in each state. A hover colour read while the pointer drifts off the element is a guess.

4. Copy the named part

Rebuild the part the user named and nothing around it. Asked for the sidebar, copying the whole layout is match-copy-the-named-part. Carry the idea, not the brand. Their logo, product names and colours stay behind unless the user asked for them. That is match-concept-not-brand.

5. Compare side by side

Put the reference and the rebuild next to each other at the same size and in the same state. Name every difference left. "Looks close" with no comparison is match-side-by-side.

Capture both in every state from step 3 at one size and one device scale factor. Crop both to the same box, aligned at the part's top-left corner. In the page's Difference view, every pixel that still lights up is a difference to name. Copy compare.html to .match/<part>/index.html, an ignored folder and never the source tree. Write compare.js beside it in the shape its header comment shows, with captures in img/, then open it.

The compare page ships with the skill, so write data only. A page with its own markup or styles is match-shipped-compare. Mark each value Match, Differs or Not Verified. The user sends what is left back with Copy Differences.

The pair

Every check is a pair. The broken line sits beside the line that replaces it.

Rule Broken Replacement
match-inspect-the-source A menu rebuilt from memory The live menu opened and inspected
match-same-platform A desktop app judged by its website The installed app opened
match-read-computed-values border-radius: 8px by eye The 6px read from computed styles
match-rendered-font font-family: "Brand Sans", Arial copied, Arial rendering in the rebuild Both sides reporting the same family from CSS.getPlatformFontsForNode
match-read-the-motion transition: transform 200ms ease guessed 240ms cubic-bezier(0.32, 0.72, 0, 1) read from getTiming() during the change
match-convert-design-units letter-spacing: -2% pasted from the design file letter-spacing: -0.02em
match-measure-at-scale gap: 24px read off a 2× screenshot gap: 12px, 24 device pixels divided by 2
match-inventory-states The resting state only Hover, pressed, open, empty and error rebuilt
match-copy-the-named-part The whole layout copied for "the sidebar" The sidebar only
match-concept-not-brand Their logo and accent colour shipped The pattern in the project's tokens
match-side-by-side "Looks close" Both at one size, differences listed
match-shipped-compare A compare page written from scratch compare.html copied, only compare.js written

Reporting

Severity. HIGH is a rebuild made without opening the source or the same platform, or a missing state the user named. MEDIUM is a guessed value under "one to one", a part copied beyond the ask or the reference's brand shipped. It is also a fallback font rendering on either side, guessed motion, a design-file unit pasted unconverted or a screenshot measured in device pixels. So is "looks close" with no side-by-side, or a hand-written compare page. LOW is a remaining difference that was found and named.

Format. One finding per root cause:

file:line - [rule-id] what is wrong

Then a Rule | Count | Severity summary. Block when any HIGH remains, Approve otherwise. Mark every value you could not read from the source as Not verified.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 warningtoday3 checks · Risk SAFE
  • Gen Agent Trust Hub1d

    The skill is a design-to-code comparison tool that analyzes external sources and generates a local visualization report. It includes standard attack surfaces for indirect prompt injection from the websites it inspects and performs local script generation to power its reporting interface.

  • Sockettoday

    No alerts

  • Snyk1d

    Risk: MEDIUM · 1 issue

Signed by skilld at b2e4283. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 12 hours ago.

Activeupdated yesterday

README badge

README badge for raphaelsalaja/skills/match