Contribute a Prompt

Every prompt in the library came from someone who needed it. Open a PR with yours.

How to contribute

1
Fork the repo

github.com/kishormorol/promptlean — fork, then clone to your machine.

2
Add your prompt to data/prompts.json

Follow the schema below. All three variants are required. Use [PLACEHOLDER] syntax for user-filled values.

3
Open a pull request

Title: feat: add "[Your Prompt Title]" to [Category]. Include the output from at least one model in the PR description.

Prompt Builder

Fill in the form — the correct JSON is generated automatically. Copy it into data/prompts.json.

0t
0t
0t
Generated JSON model_notes & benchmarks need manual completion 
{
  "id": "your-prompt-id",
  ...
}

JSON Schema

data/prompts.json 
{
  "id": "your-prompt-id",           // kebab-case, unique
  "title": "Your Prompt Title",
  "category": "Code",               // Code | Data | Research | Writing | AI/ML | Productivity | Engineering
  "tags": ["tag1", "tag2"],         // 3–6 lowercase tags
  "featured": false,                // leave false; maintainers set this
  "description": "One sentence describing what this prompt does and when to use it.",
  "model_notes": {
    "gpt-5.4": "...",
    "claude-sonnet-4-6": "...",
    "claude-opus-4-6": "...",
    "gemini-3.1-pro": "...",
    "grok-4": "...",
    "llama-4-scout": "...",
    "mistral-large-3": "...",
    "o1": "..."
  },
  "variants": {
    "lean": {
      "prompt": "Your lean prompt here. Use [PLACEHOLDER] for user content.",
      "token_estimate": 28          // count with tiktoken or similar
    },
    "balanced": {
      "prompt": "Your balanced prompt here.",
      "token_estimate": 72
    },
    "max_quality": {
      "prompt": "Your max quality prompt here.",
      "token_estimate": 165
    }
  }
}

Contribution guidelines

✓ We accept
  • Task-focused prompts with real-world use cases
  • Prompts you've actually used in production or research
  • Variants that meaningfully differ in token cost and quality
  • New categories not yet represented
  • Improvements to existing prompts (open an issue first)
✗ We decline
  • Prompts that generate harmful or deceptive content
  • Jailbreaks or prompt injection exploits
  • Vanity prompts with no clear use case
  • Duplicates of existing prompts
  • Lean variants that are just truncated max-quality prompts

Quality bar for variants

The three variants should feel genuinely different — not just shorter versions of the same prompt. The quality difference should be real and measurable.

⚡ Lean

Should produce a usable output with no wasted tokens. If the lean prompt gets 80%+ of the max-quality output, that's a good lean variant.

★ Max Quality

Should produce output that a human expert would rate noticeably higher than the lean variant. If the output is indistinguishable, the max-quality variant is not doing its job.

Community Variants

Have a specialized take on an existing prompt? Add it as a community variant — it shows up as an extra tab on the prompt page without replacing the standard ones.

How it works
  1. Fork the repo and create community/{prompt-id}.json
  2. Add your variant(s) as a JSON array (schema below)
  3. Open a PR — include example output from at least one model

Example: to add a variant for the Code Review prompt, create community/code-review.json.

community/{prompt-id}.json 
[
  {
    "id": "your-variant-id",         // kebab-case, unique within this prompt
    "label": "🔒 Your Variant Name", // shown in the tab — keep it short
    "author": "your-github-username",
    "description": "One sentence: what does this variant optimize for?",
    "prompt": "Your specialized prompt text. Use [PLACEHOLDER] for user content.",
    "token_estimate": 95             // rough count; use chars/4 as approximation
  }
]
Open GitHub →