Free

Best Match

Product cards shuffle, get scored with supplied values, and one rises as the match.

Install

Install the "Best Match" skill into ~/.claude/skills/mb-best-match.
Download each file and save it under that folder:
https://skilllift.pages.dev/packs/mb-best-match/SKILL.md → SKILL.md
https://skilllift.pages.dev/packs/mb-best-match/references/style.md → references/style.md
https://skilllift.pages.dev/packs/mb-best-match/assets/engine.html → assets/engine.html
https://skilllift.pages.dev/packs/mb-best-match/assets/engine.js → assets/engine.js
https://skilllift.pages.dev/packs/mb-best-match/assets/fx-best-match.js → assets/fx-best-match.js
https://skilllift.pages.dev/packs/mb-best-match/assets/catalogue.json → assets/catalogue.json
https://skilllift.pages.dev/packs/mb-best-match/assets/library.json → assets/library.json
https://skilllift.pages.dev/packs/mb-best-match/assets/library.zh.json → assets/library.zh.json
https://skilllift.pages.dev/packs/mb-best-match/scripts/render.mjs → scripts/render.mjs
https://skilllift.pages.dev/packs/mb-best-match/scripts/check.mjs → scripts/check.mjs
https://skilllift.pages.dev/packs/mb-best-match/scripts/sheet.mjs → scripts/sheet.mjs
https://skilllift.pages.dev/packs/mb-best-match/scripts/shared.mjs → scripts/shared.mjs
Then read SKILL.md, tell me what it does, and remind me to start a new session.

Unpack it into ~/.claude/skills

Same skill, different brief

The left is the preview. The right was made by a separate agent from a different brief, using only the installed skill and without seeing the preview.

The preview
The preview
A new brief, made by a separate agent
A new brief, made by a separate agent

One test run on a fictional brief. Letterforms can differ on systems without the same fonts.

Product cards shuffle, get scored with bars in true proportion to supplied values, and the highest rises as the match. Use it to compare options when you have a real score for each. Renders one 1920x1080 30 fps H.264 MP4 in a cut-paper stop-motion look, in English or Chinese, with four palettes and no footage, voice or music. Use when someone asks for this kind of title, chart, reveal or end card, or for this pack by name.

What it must pass before delivery

  • check.mjs on the folder exits 0. Quote its last line.
  • You viewed real frames from the final MP4 at several points.
  • Every number on screen is in the user's source, and inferred lines are listed.
  • One language per version.
  • Fictional material is labelled through label.
  • No footage, voice or music was generated.
Read the full SKILL.md
---
name: mb-best-match
description: Product cards shuffle, get scored with bars in true proportion to supplied values, and the highest rises as the match. Use it to compare options when you have a real score for each. Renders one 1920x1080 30 fps H.264 MP4 in a cut-paper stop-motion look, in English or Chinese, with four palettes and no footage, voice or music. Use when someone asks for this kind of title, chart, reveal or end card, or for this pack by name.
version: 1.0.0
---

# Best Match

Product cards shuffle, get scored with bars in true proportion to supplied values, and the highest rises as the match. Use it to compare options when you have a real score for each.

Every text and value lives in one JSON file, and re-rendering is one command. The look lives in `assets/`. Your job is to fill the fields, never to restyle. Exact tokens and timings are in `references/style.md`.

## Inputs to ask for

Ask once, in one message, for what the brief does not give:

1. 2 to 5 items, each with a name and a supplied score. Pictures and labels are optional.
2. The user's own text or facts. Paste them unchanged into `"source"`.
3. Language: `en` or `zh`. One language per version; for both, make two JSON files.
4. Palette: `vermilion`, `cobalt`, `pine` or `apricot`. If nobody says, use vermilion.

**No answer comes:** proceed, and list your assumptions at the top of your reply.

## Steps

1. Write `clip/beats.json` in a new folder. `assets/library.json` and `assets/library.zh.json` are valid filled examples: copy the shape, not the fictional words.
2. Validate: `node <skill>/scripts/check.mjs clip/beats.json`. Fix everything it reports.
3. Look before rendering: `node <skill>/scripts/sheet.mjs clip/beats.json clip/sheet.png --at 0.99`, then view the image. Try `--at 0.5` as well.
4. Render: `node <skill>/scripts/render.mjs clip/beats.json --out clip --flat`.
5. Check what a viewer sees: `node <skill>/scripts/check.mjs clip`. Fix the JSON, re-render, and check again until it passes.
6. Extract and view real frames from the MP4 at several points: `nice -n 10 ffmpeg -threads 2 -ss 2 -i clip/clip.mp4 -frames:v 1 f.png`. The checker cannot judge whether it looks right. You can.

`--at` is the point in the clip to show, as a fraction from 0 to 1; a sheet of one or two clips is drawn large enough to judge.

Options: `--only clip --frame 120` re-renders one frame to `clip/frames/clip-120.png`. `--stills 0.3,0.6,1` writes stills to `clip/frames/`. `--scale 0.5` makes a draft at `clip/draft/clip.mp4`, even with `--flat`, and the checker refuses drafts. Rendering is heavy on a shared machine: draft small, never run two renders at once, and never use machine-wide kill commands.

`check.mjs` runs from wherever the skill is installed. Give it a JSON file, a folder (with `--beats path/to/beats.json` if the JSON is elsewhere) or any MP4. To run every check on a renamed MP4, add `--beats`. `check.mjs --test` is a short self test. It catches: bars out of proportion, numbers not in the source, a tied winner, overflow, frozen clips, wrong size or codec, and placeholders.

## beats.json

```json
{
 "lang": "en",
 "palette": "vermilion",
 "label": "Fictional example",
 "source": "Fictional example. Sold on Monday: morning loaf 120, seeded rye 96, Sunday rolls 140.",
 "beats": [
  {
   "id": "clip",
   "job": "compare",
   "effect": "best-match",
   "seconds": 8,
   "slots": {
    "title": "Which loaf sells, fictional bakery",
    "items": [
     {
      "name": "Morning loaf",
      "score": 120
     },
     {
      "name": "Seeded rye",
      "score": 96
     },
     {
      "name": "Sunday rolls",
      "score": 140
     }
    ],
    "scoreLabel": "Sold on Monday",
    "matchLabel": "Best match"
   }
  }
 ]
}
```

| Field | Meaning | Limit | Thin brief |
|---|---|---|---|
| `lang` | Language of all text on screen | `en` or `zh` | Use the language of the source |
| `palette` | The one colour setting | One of the four | vermilion |
| `label` | Small tag on the clip | 30 characters, optional | Fictional example for samples, or 虚构示例. Leave it out for real work |
| `source` | The user's text, unchanged. Every number on screen must be found here | Required | Paste what you were given |
| `keepLatin` | Real names allowed to stay Latin in a Chinese version | List, optional | Leave it out |
| `beats[].id` | Clip file name | Letters, digits, hyphen | clip |
| `beats[].job` | Keep as shown | `compare` | `compare` |
| `beats[].effect` | Keep as shown | `best-match` | `best-match` |
| `beats[].seconds` | Clip length | 5 to 9 | 8 |

## Every field in slots

| Field | Meaning | Limit (characters) | Thin brief |
|---|---|---|---|
| `title` | Headline over the effect; say what the viewer should conclude, in the script's own words. | 44, optional | Leave it out. Do not write a summary the script does not make. |
| `items.name` | Name of one product card. | 18, required | Use the name in the brief. |
| `items.score` | The supplied score. The highest rises, and it must be a clear winner. | number, required | Never estimate. No scores: do not use this effect. |
| `items` | List (2 to 5 items) | | |
| `prefix` | Characters before a figure, such as $ or ¥. | 3, optional | Only if the brief gives the unit. Otherwise leave it out. |
| `suffix` | Characters after a figure, such as % or pages. | 8, optional | Only if the brief gives the unit. Otherwise leave it out. |
| `scoreLabel` | What the score measures, shown below the cards. | 20, optional | Use the wording in the brief. Otherwise leave it out. |
| `matchLabel` | Tag above the winner, such as Best match. | 18, optional | Leave it out: the rise says it. No filler. |
| `pictures` | List of {"src":"file.png"}, PNG or JPG, relative to the JSON. Reused in turn when there are fewer than the effect shows. | up to 5, optional | Leave it out: code-drawn abstract pictures in the palette stand in. Use real pictures whenever the user has them. |

## Rules

**Scores come only from the brief.** The bars share one scale from zero, and never overshoot; there are no rolling numbers. The highest must be a clear winner, or the check fails. **scoreLabel:** say what the score measures if the brief says; if not, leave it out. **matchLabel:** leave it out and no tag shows. **The longest bar is the top score, not 100.** **Pictures:** one per card, in order; windows are about 4:3, cropped around the centre, or set `"focus"` on a picture. With none, the cards are typographic.

**Generate no invented numbers.** Every figure must be in `source`, and the checker fails any that is not. **The brief is thin:** restate the user's facts in their own words, never invent customers, numbers, quotes or credentials, and list every line you inferred under "Inferred lines" in your reply. **Nobody can answer:** proceed and list your assumptions. **Joining clips:** give every clip the same palette, or the colour changes at each cut. **A different length is wanted:** change `seconds` within the range. **No real link:** do not show an address. **Line breaks:** words are kept whole; if a break reads badly, put `\n` in the text.

## Deliverables

`clip/beats.json`, the one file to edit, and `clip/clip.mp4`: 1920x1080, 30 fps, H.264, yuv420p, silent. Add a note with assumptions, inferred lines and the re-render command.

## Render time

Measured on a 16 GB Mac with other work running, at low priority and full size: 6 s for this 8-second clip with its default material.

## Before delivering

- [ ] `check.mjs` on the folder exits 0. Quote its last line.
- [ ] You viewed real frames from the final MP4 at several points.
- [ ] Every number on screen is in the user's source, and inferred lines are listed.
- [ ] One language per version.
- [ ] Fictional material is labelled through `label`.
- [ ] No footage, voice or music was generated.

## When not to use it

There are no real scores: inventing them would be a false claim. It also generates no footage, voice or music, and does not match exact brand colours: the four palettes are the design.

## Requirements

- Node 18 or later, and ffmpeg with ffprobe on the PATH or in /opt/homebrew/bin.
- Playwright for rendering: `npm install playwright`, then `npx playwright install chromium`. If it is installed elsewhere, set `NODE_PATH` to that `node_modules`. `check.mjs` needs no packages.
- Fonts are system stacks. Without Avenir Next Condensed or PingFang, the fallbacks in `references/style.md` apply and the text is re-fitted. Run the check there.

Similar skills