anti-ai-slop-design
Claude skill: workflow and prompting rules for non-generic UI, slide, and motion design
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/panyuti-ai--anti-ai-slop-design)HTML
<a href="/listings/panyuti-ai--anti-ai-slop-design"><img src="/api/listings/panyuti-ai--anti-ai-slop-design/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
Option 1: Clone and use
git clone https://github.com/panyuti-ai/anti-ai-slop-design.git cd anti-ai-slop-design claude
Inside that folder, run `/anti-ai-slop-design`. The skill lives in `.claude/skills/`, so it loads automatically for this folder only.
Option 2: Install as a plugin (any project)
/plugin marketplace add panyuti-ai/anti-ai-slop-design /plugin install anti-ai-slop-design --marketplace panyuti-ai/anti-ai-slop-design
Plugin skills are namespaced, so it appears as `/anti-ai-slop-design:anti-ai-slop-design`.
Option 3: Manual copy
Copy `.claude/skills/anti-ai-slop-design/SKILL.md` into `~/.claude/skills/anti-ai-slop-design/SKILL.md` to use it in every project without a plugin.
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 11 lines
- {
- "name": "anti-ai-slop-design",
- "owner": { "name": "panyuti-ai" },
- "plugins": [
- {
- "name": "anti-ai-slop-design",
- "source": "./",
- "description": "Workflow and prompting rules for getting non-generic UI, slide, and motion design out of Claude."
- }
- ]
- }
README.md · README · 216 lines
- <div align="center">
- # anti-ai-slop-design
- **Get distinctive UI, slide, and motion design out of Claude, not the same serif-and-orange look every time.**
- 
- 
- 
- [Install](#install) · [The method](#the-method) · [Prompts to copy](#prompts-to-copy) · [Checklist](#checklist) · [Source](#source-and-attribution)
- </div>
- ---
- ## The idea in one line
- > What you put in is what you get out. Give context, ask for breadth, then converge.
- Claude's default design tends to repeat itself: serif text, light gray background, orange accent, cards with a colored edge stripe. This skill is a working method for steering around that.
- ## Install
- ### Option 1: Clone and use
- ```bash
- git clone https://github.com/panyuti-ai/anti-ai-slop-design.git
- cd anti-ai-slop-design
- claude
- ```
- Inside that folder, run `/anti-ai-slop-design`. The skill lives in `.claude/skills/`, so it loads automatically for this folder only.
- ### Option 2: Install as a plugin (any project)
- ```text
- /plugin marketplace add panyuti-ai/anti-ai-slop-design
- /plugin install anti-ai-slop-design --marketplace panyuti-ai/anti-ai-slop-design
- ```
- Plugin skills are namespaced, so it appears as `/anti-ai-slop-design:anti-ai-slop-design`.
- ### Option 3: Manual copy
- Copy `.claude/skills/anti-ai-slop-design/SKILL.md` into `~/.claude/skills/anti-ai-slop-design/SKILL.md` to use it in every project without a plugin.
- ## The method
- Open each section to read the full rule. The same content lives in [`SKILL.md`](.claude/skills/anti-ai-slop-design/SKILL.md), which is what Claude loads.
- <details open>
- <summary><b>1. Start from context, not a blank request</b></summary>
- - Collect inputs before generating: screenshots as a mood board, pages that feel right, past work you like, and a short statement of what the thing is and how it should feel.
- - If a codebase exists, read it first (see step 4).
- </details>
- <details>
- <summary><b>2. Diverge, then converge</b></summary>
- 1. Ask for about 10 options that differ in fonts, colors, and visual style.
- 2. Pick pieces: "I like part of #1 and part of #3. Make 5 more that are different but built on those."
- 3. Repeat until one direction is clear.
- 4. Keep every version side by side on the canvas. Copy a version before changing it instead of overwriting.
- </details>
- <details>
- <summary><b>3. Wireframe first when the task is hard</b></summary>
- - Ask for several wireframes with different layouts before any font or color decision.
- - Ask Claude to name similar existing products, compare their layouts, and pick the one users already know how to use.
- </details>
- <details>
- <summary><b>4. Existing product: recreate from code</b></summary>
- - Tell Claude to read the relevant parts of the codebase and recreate the current state on the canvas, then produce variations from there.
- - Do not maintain long-lived design files. Recreate from current code each time.
- - To build, continue in the same session so styles, sizes, fonts, and colors carry over without translation loss.
- </details>
- <details>
- <summary><b>5. Decide style before producing a full deck</b></summary>
- For presentations, agree on the visual style first. Redoing a whole deck in the wrong style wastes tokens.
- </details>
- <details>
- <summary><b>6. Block Claude's default look</b></summary>
- - Name Claude's habits in the prompt and forbid them unless asked.
- - Better: supply the brand or identity yourself (colors, fonts, tone). Brand context is a durable asset that works across models.
- - If output feels corporate, say so and push: "Too corporate. Use full creative range. Make it feel like a film, not a slide deck."
- </details>
- <details>
- <summary><b>7. Give feedback efficiently</b></summary>
- - High-level direction works ("make the card list denser, no subheadings, no dividers"), as do small specifics ("reduce left padding", "align icons to cells").
- - Voice notes are fine for loose thoughts.
- - Use comments pinned to an element or a video timestamp instead of describing location in words. Then ask Claude to process all comments.
- - Let Claude ask about 8 clarifying questions before designing. Answering is easier than guessing what a designer would ask. Allow skipping.
- </details>
- <details>
- <summary><b>8. Motion and animation</b></summary>
- - Claude can track many things at once; people cannot. Require one focal change at a time.
- - Leave reading time: when text appears, ask whether people have enough time to read it before the next element, and whether their eyes are where the next action happens.
- - If it looks like static slides, ask for continuous movement, such as a single line the eye follows.
- - Process: storyboard (static, cheap) → a few visual style options → pick one → iterate → comment-based polish.
- </details>
- <details>
- <summary><b>9. Dashboards</b></summary>
- - Give Claude the data source (spreadsheet link, MCP connector) and ask it to "make this a dashboard". Leaving the metrics unspecified can produce useful surprises.
- - Connect live data when the source supports it.
- - Check numbers against their source: inspect the code Claude used to fetch each figure.
- </details>
- <details>
- <summary><b>10. Evaluate results</b></summary>
- - Prefer yes/no checks over 1-5 scores.
- - Automate only basics: text overlap, and whether 10 runs all pick the same font and color.
- - Taste still needs a human reviewing outputs and transcripts.
- - To test a skill that produces videos or designs, run several fresh sessions that see only the skill plus a one-line request, then review the results yourself.
- </details>
- ## Prompts to copy
- <details>
- <summary><b>Diverge</b></summary>
- ```text
- Here is what we're designing and how it should feel: <description>.
- Reference screenshots attached. Give me 10 options that differ in fonts,
- colors, and visual style.
- ```
- </details>
- <details>
- <summary><b>Converge</b></summary>
- ```text
- None is perfect. I like <part> of #1 and <part> of #3. Make 5 more that
- are different from each other but built on those.
- ```
- </details>
- <details>
- <summary><b>Recreate from code</b></summary>
- ```text
- Read the relevant parts of the codebase, recreate the current <screen> on
- the canvas, then give me variations. Keep the original next to them.
- ```
- </details>
- <details>
- <summary><b>Block the defaults</b></summary>
- ```text
- You tend to overuse the same two fonts and two colors (serif text, light
- gray background, orange accent, card edge stripes). Do not use them unless
- I ask or it is clearly right for this design.
- ```
- </details>
- <details>
- <summary><b>Motion pacing</b></summary>
- ```text
- People can only focus on one thing at a time and read slowly. Make one
- focal change at a time, and leave enough time for text to be read before
- the next element appears.
- ```
- </details>
- ## Checklist
- - [ ] Context supplied (mood board, references, codebase)
- - [ ] 10 divergent options requested, then a converged round
- - [ ] Wireframes first if the layout is the hard part
- - [ ] Default fonts and colors explicitly excluded, or brand supplied
- - [ ] Versions kept side by side
- - [ ] Motion: one focal change at a time, reading time left
- - [ ] Data figures traced to source
- - [ ] A human has looked at the output
- ## Source and attribution
- This is an independent summary of practices discussed in the video "Full Tutorial: How to Fix AI Slop with Claude Design in 40 Minutes" (Peter Yang interviewing Nate Parrott): https://www.youtube.com/watch?v=604T-74s3KE
- It contains no transcript text. It is not affiliated with or endorsed by the interviewer, the guest, or Anthropic. Any errors in interpretation are mine.
- ## License
- [MIT](LICENSE)
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.