How to use the slide-craft agent skill for slide editing

What is an agent skill, and what does slide-craft make the agent do?

An agent skill is a folder with a SKILL.md file that an AI agent loads when a task matches the skill's description. slide-craft is a reference guide for the agent, not a standalone tool. It describes the design law of a slide page: the canvas geometry, the text height table, the type scale, the contrast pairs, the spacing rhythm, and which element type carries which content.

When the agent changes one element on a slide, such as its position, size, text, colour or contrast, the skill has it check the change against those rules so the edited slide still looks consistent. The skill itself says it is the standard edit reference for the page-clone and pro-editing skills, which decide which pages to touch and in what order.

• Keeps elements inside a 1000 x 562.5 canvas with a 50px margin
• Sizes text boxes from a height table instead of guessing
• Picks font sizes from a fixed type scale
• Checks colour changes against a contrast target
• Writes rich text in markup the renderer actually displays

Install slide-craft from the ZIP

Download the ZIP for slide-craft from the directory page. The skill ships as a single file, SKILL.md. Unzip it so that the skill folder ends up in one of the locations below.

For a personal setup, put it in ~/.claude/skills/. For a single project, put it in the project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/slide-craft/
• Project: .claude/skills/slide-craft/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Which requests trigger the skill?

The skill loads when the task matches its description. You do not need to name it. Ask for the kind of change you want on a slide, and the agent is likely to load it when the work involves one of the situations the skill lists: patching elements one at a time, a colour or contrast fix, text that must grow or shrink inside a box, a page that feels crowded or empty, or replacement content that needs real rich-text structure.

Some natural requests that match these situations:

• Make the subtitle on slide 4 darker, it is hard to read on the blue panel.
• The text on slide 7 overflows its box. Fix it without changing the other elements.
• Slide 2 feels crowded. Adjust the spacing and sizes so it breathes.
• Rewrite the three bullet points on slide 5 and keep the markup valid for the renderer.
• Move the chart caption on slide 3 so it lines up with the chart's left edge.

Understand the main rules

Text boxes have 10px padding on all four sides, so the usable area is (width - 20) x (height - 20). The skill gives a lookup table of heights for each font size from 14px to 36px and for 1 to 5 lines, with line-height 1.5 and padding included.

When you replace words, the skill tells the agent to work out characters per line as (width - 20) / font size, keep the longest line at 75% of that or less, count the lines the new content needs, add about 0.8 of a line of slack, and look up the height against the largest font size in the content. If text overflows, the fix order is shorten the words, then step to the next table row, then widen the box.

Colour changes follow the same logic. Before changing a colour, identify what the text sits on. The skill sets a target of at least 4.5:1 for body text and at least 3:1 for titles at 24px and above.

• Type scale: main title 32-36px, subtitle 24-28px, key points 18-20px, body 16-18px, caption 14-16px
• Alignment: there is no alignment operation, so the agent writes each coordinate itself
• Centred content: left = (1000 - width) / 2, recomputed each time
• Right-aligned content: left = 1000 - width - 60

Know how the renderer treats HTML

Text content is HTML, and the skill lists renderer quirks that affect how rich text must be written. Lists written as ul and li render without markers, so the skill recommends one p per item with the bullet character in the text. Heading tags h1 to h6 are not styled differently, so size and weight should come from inline font-size and strong.

Stacked p tags have no gap between them, and a newline in plain text collapses to a space. Supported inline styles are font-size, color, text-align, line-height, font-weight and font-family. Supported tags are p, span, strong or b, em or i, and u.

• Bullets: one p per item, for example <p style="font-size:18px;">• First point</p>
• Multi-line text: use real p tags, one per line
• Literal less-than signs in prose must be sent as escaped entities