Contribute a prompt

Prompts here help people build, ship and run small AI-powered web apps: specs, in-app AI features, Claude Code workflows, design systems, deployment, QA and data tooling. One strong, tested entry beats five sketches.

Submit one

1. Fork the repo and create a branch.
2. Copy TEMPLATE.md to content/<category>/<slug>.md. The slug is the file name: lowercase words joined by hyphens, unique across the library.
3. Fill in the frontmatter and the four sections below.
4. If the entry is a reusable workflow, also add skills/<slug>/SKILL.md (see Skills).
5. Run npm ci && npm run validate and fix anything it reports. npm run dev builds the site and serves it locally.
6. Open a pull request. The PR template has a short checklist, and CI runs the same validation.

Frontmatter

---
title: Ask 2–4 clarifying questions with tappable options   # 3–90 characters
summary: One sentence for the card, 20–200 characters.
category: ai-features        # must match the folder
tags: [structured-output, ux] # 1–6, lowercase-hyphenated, 24 characters at most; reuse tags already on the site
model_tested: Claude Opus 5.5 # "Claude <model>", 8–40 characters: the model you ran this exact text on
date_tested: 2026-11-06       # YYYY-MM-DD, the UTC day you ran it; not in the future, not before 2023
difficulty: beginner          # beginner | intermediate | advanced
use_as: prompt                # prompt | skill | claude-code
credit: Optional, one line    # e.g. the idea or source you adapted
---

Field	Meaning
category	One of app-specs, ai-features, claude-code, design, deploy-hosting, qa-debugging, data-geospatial. Propose a new one in your PR if none fits.
use_as	prompt: paste into a Claude chat or call it from code. skill: install it as a skill (needs a SKILL.md). claude-code: run it in a Claude Code session on a repository.
model_tested, date_tested	The model and day you last ran this exact text and got the output in your example. The site flags entries older than six months as "may be dated".
difficulty	How much context the reader needs to use it well, not how long the prompt is.

No other fields are allowed; the validator says which one it doesn't know.

Sections

Use exactly these ## headings, in this order:

1. ## The prompt: the prompt itself in a fenced code block. The site's Copy button copies the first fenced block in this section. Mark the parts a reader must fill in as {{snake_case}} placeholders; the site renders an input box for each one. A sentence before the block saying where to run it is welcome.
2. ## Example: a short, real input and the output it produced, trimmed. Real means you ran it. Don't write the output yourself.
3. ## Why it works: 2–4 points on the design choices: what each rule prevents or enables.
4. ## Variations (optional): adaptations, and where the idea came from.

Don't add a # heading; the title comes from the frontmatter.

Skills

If the entry is a workflow Claude should follow every time rather than a one-off prompt, add a skill:

skills/<slug>/SKILL.md          # required: frontmatter with name: <slug> and a description
skills/<slug>/references/*.md   # optional: files the skill loads on demand

The description says when Claude should use the skill, in 20–1024 characters. The entry page links to it, and readers download it into .claude/skills/<slug>/ or upload it in the Claude apps under Customize → Skills.

Quality bar

• Tested. You ran this text, on the model you name, and the example is real output.
• General. Replace anything specific to your project with a placeholder, unless it is the example.
• Explained. "Why it works" teaches the technique, so readers can adapt it rather than copy it blind.
• Short enough. Cut any rule you can't explain in "Why it works".

Keep it clean

Pull requests are public. Don't include:

• API keys, tokens or passwords. The validator rejects common key formats.
• Links to private systems, private repositories or private chat or session links.
• Names or details of real people who aren't public figures, or of your employer, clients or their sites.
• Confidential business information.

Swap real details for neutral ones before you commit; git history is public too.

Licences

By contributing you agree that your prompt content (content/, skills/) is published under CC BY 4.0 and anything else under MIT.