Skip to main content

Web Prompts

Shareable, versionable prompt templates. Agent-ready prompts as web resources.

The Problem​

Today's prompts are:

  • Copy-pasted — No version control
  • Siloed — Locked in apps
  • Opaque — Users don't know what they're running
  • Incompatible — Different formats per platform

Web Prompts makes prompts first-class web citizens.

Core Concepts​

Prompts as URLs​

Every prompt has a URL:

https://prompts.example/writing/blog-post/v2.1

Fetch it like any web resource:

const response = await fetch('https://prompts.example/writing/blog-post/v2.1');
const prompt = await response.json();

Structured Format​

{
  "@context": "https://webprompts.org/context.jsonld",
  "@type": "Prompt",
  "id": "https://prompts.example/writing/blog-post/v2.1",
  "name": "Blog Post Writer",
  "description": "Creates engaging blog posts from outlines",
  "version": "2.1",
  "author": "https://alice.example/profile#me",
  "template": "Write a blog post about {{topic}}...",
  "parameters": [
    {
      "name": "topic",
      "type": "string",
      "required": true
    },
    {
      "name": "tone",
      "type": "enum",
      "values": ["casual", "professional", "academic"],
      "default": "casual"
    }
  ],
  "tags": ["writing", "content", "blog"]
}

Versioning​

Semantic versioning for prompts:

  • v1.0 → v1.1 — Minor refinements
  • v1.1 → v2.0 — Major changes to behavior
  • latest — Always the newest version

Composition​

Prompts can extend other prompts:

{
  "@type": "Prompt",
  "extends": "https://prompts.example/writing/base",
  "name": "Technical Blog Post",
  "template": "{{base.template}}\n\nFocus on technical accuracy..."
}

Discovery​

Registries​

Discover prompts through registries:

GET https://registry.webprompts.org/search?tags=writing&sort=popularity

In Your Pod​

Store personal prompts in your Solid pod:

https://alice.pod/prompts/
├── work/
│   ├── email-reply.json
│   └── meeting-summary.json
└── personal/
    ├── journal-entry.json
    └── recipe-generator.json

Agent Integration​

MCP Tools​

Web Prompts integrates with MCP:

{
  "name": "execute_prompt",
  "description": "Execute a web prompt",
  "inputSchema": {
    "type": "object",
    "properties": {
      "promptUrl": { "type": "string" },
      "parameters": { "type": "object" }
    }
  }
}

Example Usage​

// Fetch and execute a prompt
const prompt = await fetch('https://prompts.example/coding/review').then(r => r.json());

const result = await agent.execute({
  prompt: prompt.template,
  variables: {
    code: myCode,
    language: 'javascript'
  }
});

Trust and Safety​

Signed Prompts​

Prompts can be cryptographically signed:

{
  "@type": "Prompt",
  "signature": {
    "type": "Ed25519Signature2020",
    "created": "2025-01-19T12:00:00Z",
    "verificationMethod": "did:nostr:npub1...",
    "proofValue": "..."
  }
}

Content Policies​

Registries can enforce content policies:

  • No malicious instructions
  • No jailbreak attempts
  • Required safety guidelines

Benefits​

TraditionalWeb Prompts
Copy-paste chaosVersion controlled
Platform-lockedPortable across tools
No attributionAuthor linked
Can't verifyCryptographically signed
Siloed discoveryRegistry-based search

See Also​