The Prompt Package Manager (PRPM) provides a universal registry and package management system for AI coding tools, including prompts, agents, skills, and rules. It is designed to address the fragmentation of formats across different AI-powered IDEs and assistants like Cursor, Claude, Continue, and GitHub Copilot. The core philosophy is "Write Once, Distribute Everywhere," allowing authors to publish a package in a single format, which can then be installed and used in any supported editor through automatic, on-the-fly format conversion.

This system enables developers to discover, share, version, and manage AI development artifacts with an experience similar to traditional package managers like npm. It supports both individual packages and "Collections," which are bundles of related packages for complete workflow setups.

Sources: packages/webapp/src/app/blog/introducing-prpm/page.tsx:29-43, packages/webapp/src/app/blog/write-once-distribute-everywhere/page.tsx:40-47, packages/webapp/public/blog/prpm-complete-feature-guide.md:10-21

Package Lifecycle

The lifecycle of a PRPM package follows standard package management practices, from creation to consumption.

Generating architecture diagram...

This diagram illustrates the typical workflow for a package author and consumer within the PRPM ecosystem.
Sources: CONTRIBUTING.md:12-18, packages/webapp/public/blog/prpm-complete-feature-guide.md:330-349

Initialization: A new package is scaffolded using prpm init. This command interactively creates a package.prpm.json manifest and can generate example files based on the chosen format. It also features a "smart init" mode that can detect existing package files and reconcile them with a manifest.
Sources: packages/cli/src/commands/init.ts:18-28, 410-413
Publishing: Once the package content is ready, the author publishes it to the PRPM registry using prpm publish.
Sources: CONTRIBUTING.md:18
Discovery: Users can find packages through the web interface or via the CLI using prpm search (keyword-based) or prpm ai-search (semantic, intent-based).
Sources: packages/webapp/public/blog/prpm-complete-feature-guide.md:50-65
Installation: A user installs a package with prpm install <package-name>. The CLI automatically detects the user's editor and converts the package to the appropriate format.
Sources: packages/webapp/public/blog/prpm-complete-feature-guide.md:10-21

Package Manifest

The package.prpm.json file is the manifest that defines a package's metadata. The prpm init command assists in its creation.

Sources: CONTRIBUTING.md:21, packages/cli/src/commands/init.ts:18-28

Field	Type	Description	Source
`name`	`string`	The unique name of the package.	CONTRIBUTING.md:24
`version`	`string`	The semantic version of the package.	CONTRIBUTING.md:25
`description`	`string`	A brief summary of what the package does.	CONTRIBUTING.md:26
`format`	`string`	The native format of the package (e.g., `cursor`, `claude`, `agents.md`).	packages/cli/src/commands/init.ts:51
`subtype`	`string`	The specific type within the format (e.g., `rule`, `skill`, `agent`).	packages/cli/src/commands/init.ts:52
`category`	`string`	The category the package belongs to (e.g., `code-quality`, `testing`).	CONTRIBUTING.md:28
`tags`	`string[]`	An array of keywords to aid discovery.	CONTRIBUTING.md:29
`author`	`string`	The name and/or email of the package author.	CONTRIBUTING.md:30
`license`	`string`	The software license for the package (e.g., "MIT").	CONTRIBUTING.md:31
`files`	`string[]`	An array of file paths that constitute the package content.	packages/cli/src/commands/init.ts:32-33

Example `package.prpm.json`

{
  "name": "my-awesome-rule",
  "version": "1.0.0",
  "description": "A helpful Cursor rule for...",
  "type": "cursor",
  "category": "code-quality",
  "tags": ["typescript", "testing", "best-practices"],
  "author": "your-username",
  "license": "MIT",
  "content": "./rule.md"
}

Sources: CONTRIBUTING.md:23-34

Package Formats and Types

PRPM is designed to be format-agnostic, supporting a wide range of formats used by different AI coding assistants. This is a cornerstone of the "Write Once, Distribute Everywhere" philosophy.

Sources: packages/webapp/src/app/blog/write-once-distribute-everywhere/page.tsx:40-47

Supported Formats

The system supports conversion between 8 major formats, each with its own structure and conventions.

Format	File Location(s)	Frontmatter	Key Features
Cursor	`.cursor/rules`	Required	4 rule types, MDC format
Claude Code	`.claude/{agents,skills,commands}/`	Required	`allowed-tools`, model selection, hooks
Continue	`.continue/rules/*.md`	Required	Globs, regex matching, `alwaysApply` logic
Windsurf	`.windsurf/rules`	None	Plain markdown, 12k character limit
GitHub Copilot	`.github/copilot-instructions.md`	Optional	Two-tier system (repo-wide, path-specific)
Kiro Steering	`.kiro/steering/*.md`	Optional	Inclusion modes, foundational types
Kiro Hooks	`.kiro/hooks/*.json`	N/A (JSON)	Event-driven file automations
agents.md	`agents.md`	None	Open standard, plain markdown

Sources: packages/webapp/src/app/blog/format-specifications-guide/page.tsx:102-167

Package Types

Packages can be categorized by their function, such as rule, agent, skill, prompt, and more. These types help define the package's purpose.

Sources: CONTRIBUTING.md:36-43

Format Conversion System

A core feature of PRPM is its ability to perform "lossless, semantically-aware conversions" between formats. When a user installs a package, PRPM converts it from its source format to the format required by the user's target editor.

Generating architecture diagram...

This diagram shows how a package published in one format is converted to another upon installation.
Sources: packages/webapp/src/app/blog/write-once-distribute-everywhere/page.tsx:54-60, packages/webapp/src/app/blog/format-specifications-guide/page.tsx:610-622

The conversion process involves:

Parsing the source package into a Canonical Format.
Mapping equivalent capabilities, such as file pattern matching, across formats.
Preserving format-specific features where possible, often by adding them as comments or documentation in the converted output.
Validating the output against the target format's JSON schema to ensure correctness.

Sources: packages/webapp/src/app/blog/write-once-distribute-everywhere/page.tsx:120-150, packages/webapp/src/app/blog/format-specifications-guide/page.tsx:612-622

The `agents.md` Standard

agents.md is an open, tool-agnostic format for providing project-specific guidance to AI agents. It was created through a collaboration involving OpenAI and Google. Its design philosophy is "radical simplicity."

Key characteristics:

Single File: A single agents.md file at the project root.
Plain Markdown: No special syntax, extensions, or required frontmatter.
Human-First: Designed to be as readable as standard documentation.
Tool-Agnostic: Supported by multiple tools, including OpenAI Codex, GitHub Copilot, and Google Gemini Code Assist, avoiding vendor lock-in.

While there are no required fields, recommended sections include:

Project Overview
Tech Stack
Development Setup
Code Style Guidelines
Testing Instructions
Security Practices

GitHub Copilot supports agents.md in addition to its native .github/copilot-instructions.md, and will merge instructions from both if they exist.

Sources: packages/webapp/src/app/blog/agents-md-deep-dive/page.tsx:40-46, 61-80, 150-175, packages/webapp/src/app/blog/copilot-deep-dive/page.tsx:55-66

Collections

Collections are bundles of related packages that can be installed with a single command. They are designed to provide complete setups for specific workflows or technology stacks (e.g., a "Next.js Professional Setup").

A collection is defined by a collection.prpm.json manifest, which lists the package IDs to be included.

Example `collection.prpm.json`

{
  "scope": "@collection",
  "id": "nextjs-pro",
  "name": "Next.js Professional Setup",
  "version": "1.0.0",
  "description": "Complete Next.js development environment",
  "category": "frontend",
  "packages": [
    {
      "id": "react-best-practices",
      "version": "^1.0.0",
      "required": true
    },
    {
      "id": "typescript-strict",
      "version": "^1.0.0",
      "required": true
    }
  ]
}

Sources: CONTRIBUTING.md:69-93, packages/webapp/public/blog/prpm-complete-feature-guide.md:78-90

Summary

The PRPM package management system provides the foundational infrastructure for a cross-platform ecosystem of AI coding tools. By abstracting away format differences through a canonical representation and on-the-fly conversion, it allows authors to reach a wider audience and enables users to access a larger pool of tools, regardless of their preferred AI assistant. Key components like the package manifest, collections, and support for open standards like agents.md create a robust and flexible system for distributing AI-executable knowledge.

Sources: packages/webapp/src/app/blog/distributable-intelligence/page.tsx:40-50, packages/webapp/src/app/blog/introducing-prpm/page.tsx:46-51

Resources

AI code review
Bug Bench
AI coding statistics
Coding standards
Agents
Wikis

View on GitHub

Back to directory

Repository

pr-pm/prpm

Overview

Welcome to the PRPM Wiki Getting Started Contributing Guide

System Architecture

Architecture Overview Monorepo Structure Shared Types

Core Features

Extensibility and Customization

Prompt Format Conversion Engine Supported AI Tool Formats Creating Custom Hooks

Backend Systems (Registry)

Registry API Overview API Authentication and Security Package Management API Search and Discovery API User and Organization API Payments and Subscriptions API Background Jobs and Cron Scheduler

Data Management

Database Schema Database Migrations Package Scoring and Analytics Embedding Generation for AI Search

Frontend Components (Webapp)

Webapp Overview Package Discovery UI AI Prompt Playground User Dashboard and Analytics UI

Deployment and Infrastructure

Local Development Setup Testing Strategy Deployment and CI/CD

Package Management API

Relevant source files

The following files were used as context for generating this wiki page:

packages/cli/src/commands/init.ts
CONTRIBUTING.md
packages/webapp/src/app/blog/format-specifications-guide/page.tsx
packages/webapp/src/app/blog/agents-md-deep-dive/page.tsx
packages/webapp/public/blog/prpm-complete-feature-guide.md
packages/webapp/src/app/blog/write-once-distribute-everywhere/page.tsx
AGENTS.md
packages/webapp/src/app/blog/copilot-deep-dive/page.tsx

Package Lifecycle

The lifecycle of a PRPM package follows standard package management practices, from creation to consumption.

Generating architecture diagram...

Initialization: A new package is scaffolded using prpm init. This command interactively creates a package.prpm.json manifest and can generate example files based on the chosen format. It also features a "smart init" mode that can detect existing package files and reconcile them with a manifest.
Sources: packages/cli/src/commands/init.ts:18-28, 410-413
Publishing: Once the package content is ready, the author publishes it to the PRPM registry using prpm publish.
Sources: CONTRIBUTING.md:18
Discovery: Users can find packages through the web interface or via the CLI using prpm search (keyword-based) or prpm ai-search (semantic, intent-based).
Sources: packages/webapp/public/blog/prpm-complete-feature-guide.md:50-65
Installation: A user installs a package with prpm install <package-name>. The CLI automatically detects the user's editor and converts the package to the appropriate format.
Sources: packages/webapp/public/blog/prpm-complete-feature-guide.md:10-21

Package Manifest

The package.prpm.json file is the manifest that defines a package's metadata. The prpm init command assists in its creation.

Sources: CONTRIBUTING.md:21, packages/cli/src/commands/init.ts:18-28

Field	Type	Description	Source
`name`	`string`	The unique name of the package.	CONTRIBUTING.md:24
`version`	`string`	The semantic version of the package.	CONTRIBUTING.md:25
`description`	`string`	A brief summary of what the package does.	CONTRIBUTING.md:26
`format`	`string`	The native format of the package (e.g., `cursor`, `claude`, `agents.md`).	packages/cli/src/commands/init.ts:51
`subtype`	`string`	The specific type within the format (e.g., `rule`, `skill`, `agent`).	packages/cli/src/commands/init.ts:52
`category`	`string`	The category the package belongs to (e.g., `code-quality`, `testing`).	CONTRIBUTING.md:28
`tags`	`string[]`	An array of keywords to aid discovery.	CONTRIBUTING.md:29
`author`	`string`	The name and/or email of the package author.	CONTRIBUTING.md:30
`license`	`string`	The software license for the package (e.g., "MIT").	CONTRIBUTING.md:31
`files`	`string[]`	An array of file paths that constitute the package content.	packages/cli/src/commands/init.ts:32-33

Example `package.prpm.json`

{
  "name": "my-awesome-rule",
  "version": "1.0.0",
  "description": "A helpful Cursor rule for...",
  "type": "cursor",
  "category": "code-quality",
  "tags": ["typescript", "testing", "best-practices"],
  "author": "your-username",
  "license": "MIT",
  "content": "./rule.md"
}

Sources: CONTRIBUTING.md:23-34

Package Formats and Types

PRPM is designed to be format-agnostic, supporting a wide range of formats used by different AI coding assistants. This is a cornerstone of the "Write Once, Distribute Everywhere" philosophy.

Sources: packages/webapp/src/app/blog/write-once-distribute-everywhere/page.tsx:40-47

Supported Formats

The system supports conversion between 8 major formats, each with its own structure and conventions.

Format	File Location(s)	Frontmatter	Key Features
Cursor	`.cursor/rules`	Required	4 rule types, MDC format
Claude Code	`.claude/{agents,skills,commands}/`	Required	`allowed-tools`, model selection, hooks
Continue	`.continue/rules/*.md`	Required	Globs, regex matching, `alwaysApply` logic
Windsurf	`.windsurf/rules`	None	Plain markdown, 12k character limit
GitHub Copilot	`.github/copilot-instructions.md`	Optional	Two-tier system (repo-wide, path-specific)
Kiro Steering	`.kiro/steering/*.md`	Optional	Inclusion modes, foundational types
Kiro Hooks	`.kiro/hooks/*.json`	N/A (JSON)	Event-driven file automations
agents.md	`agents.md`	None	Open standard, plain markdown

Sources: packages/webapp/src/app/blog/format-specifications-guide/page.tsx:102-167

Package Types

Packages can be categorized by their function, such as rule, agent, skill, prompt, and more. These types help define the package's purpose.

Sources: CONTRIBUTING.md:36-43

Format Conversion System

Generating architecture diagram...

The conversion process involves:

Parsing the source package into a Canonical Format.
Mapping equivalent capabilities, such as file pattern matching, across formats.
Preserving format-specific features where possible, often by adding them as comments or documentation in the converted output.
Validating the output against the target format's JSON schema to ensure correctness.

Sources: packages/webapp/src/app/blog/write-once-distribute-everywhere/page.tsx:120-150, packages/webapp/src/app/blog/format-specifications-guide/page.tsx:612-622

The `agents.md` Standard

Key characteristics:

Single File: A single agents.md file at the project root.
Plain Markdown: No special syntax, extensions, or required frontmatter.
Human-First: Designed to be as readable as standard documentation.
Tool-Agnostic: Supported by multiple tools, including OpenAI Codex, GitHub Copilot, and Google Gemini Code Assist, avoiding vendor lock-in.

While there are no required fields, recommended sections include:

Project Overview
Tech Stack
Development Setup
Code Style Guidelines
Testing Instructions
Security Practices

GitHub Copilot supports agents.md in addition to its native .github/copilot-instructions.md, and will merge instructions from both if they exist.

Sources: packages/webapp/src/app/blog/agents-md-deep-dive/page.tsx:40-46, 61-80, 150-175, packages/webapp/src/app/blog/copilot-deep-dive/page.tsx:55-66

Collections

A collection is defined by a collection.prpm.json manifest, which lists the package IDs to be included.

Example `collection.prpm.json`

{
  "scope": "@collection",
  "id": "nextjs-pro",
  "name": "Next.js Professional Setup",
  "version": "1.0.0",
  "description": "Complete Next.js development environment",
  "category": "frontend",
  "packages": [
    {
      "id": "react-best-practices",
      "version": "^1.0.0",
      "required": true
    },
    {
      "id": "typescript-strict",
      "version": "^1.0.0",
      "required": true
    }
  ]
}

Sources: CONTRIBUTING.md:69-93, packages/webapp/public/blog/prpm-complete-feature-guide.md:78-90

Summary

Sources: packages/webapp/src/app/blog/distributable-intelligence/page.tsx:40-50, packages/webapp/src/app/blog/introducing-prpm/page.tsx:46-51

Resources

AI code review
Bug Bench
AI coding statistics
Coding standards
Agents
Wikis

View on GitHub

Package Management API

Package Lifecycle

Package Manifest

Example package.prpm.json

Package Formats and Types

Supported Formats

Package Types

Format Conversion System

The agents.md Standard

Collections

Example collection.prpm.json

Summary

Package Management API

Package Lifecycle

Package Manifest

Example package.prpm.json

Package Formats and Types

Supported Formats

Package Types

Format Conversion System

The agents.md Standard

Collections

Example collection.prpm.json

Summary

Example `package.prpm.json`

The `agents.md` Standard

Example `collection.prpm.json`

Example `package.prpm.json`

The `agents.md` Standard

Example `collection.prpm.json`