Overview
Meta packages are packages that teach AI how to create other packages. They’re “prompts about writing prompts” - expert guidance that helps AI assistants generate high-quality, well-structured PRPM packages.
Think of them as:
- Package creation templates
- Style guides for AI prompts
- Best practices for package structure
- Quality standards enforcement
AI assistants are great at generating content, but they need guidance to produce consistently high-quality packages that follow PRPM conventions.
Without meta packages:
- ❌ Inconsistent package structures
- ❌ Poor CSO (Context Search Optimization)
- ❌ Missing required fields
- ❌ Unclear when to use packages
- ❌ No real-world examples
With meta packages:
- ✅ Standardized structure across all packages
- ✅ Optimized for AI discovery and use
- ✅ Complete metadata and documentation
- ✅ Clear triggering conditions
- ✅ Real examples and patterns
PRPM uses meta packages to maintain consistency across its own repository:
1. Creating Skills (@prpm/creating-skills-skill)
Teaches AI how to create Claude Code skills:
prpm install @prpm/creating-skills-skill
This package exists in PRPM’s own repo at .claude/skills/creating-skills/SKILL.md and is used to generate all Claude skills in the project.
- Proper frontmatter structure
- CSO (Context Search Optimization) keywords
- When to trigger the skill
- Real-world examples
- File organization
Example output:
---
name: postgres-migrations
description: Use when working with PostgreSQL migrations or encountering migration errors
---
# PostgreSQL Migrations Expert
## When to Use
- Writing new migration files
- Debugging migration errors
- Adding indexes or generated columns
- Working with full-text search
## Common Patterns
[Real, tested examples...]
2. Creating Cursor Rules (@prpm/creating-cursor-rules-skill)
Teaches AI how to create Cursor IDE rules:
# For Claude Code
prpm install @prpm/creating-cursor-rules-skill
# For Cursor IDE
prpm install @prpm/creating-cursor-rules
Claude version: .claude/skills/creating-cursor-rules-skill/SKILL.md
Cursor version: .cursor/rules/creating-cursor-rules.mdc
.mdc format structure- Frontmatter with
ruleType and alwaysApply
- Conditional vs always-applied rules
- Best practices for rule scoping
3. Slash Command Builder (@prpm/slash-command-builder-skill)
Teaches AI how to create Claude Code slash commands:
prpm install @prpm/slash-command-builder-skill
This package includes multiple reference files:
SKILL.md - Main guidanceEXAMPLES.md - Real command examplesFRONTMATTER.md - Frontmatter specificationsPATTERNS.md - Common patterns
- Command frontmatter structure
- Argument and option definitions
- Tool permissions and model selection
- Error handling patterns
- Testing strategies
4. Agent Builder (@prpm/agent-builder-skill)
Teaches AI how to create Claude Code subagents:
prpm install @prpm/agent-builder-skill
Package location: .claude/skills/agent-builder/ with SKILL.md and EXAMPLES.md
- Agent system prompt design
- Tool access configuration
- Model selection (Sonnet vs Haiku)
- Delegation patterns
- Agent lifecycle management
Example: Package Creator for Python
Let’s create a meta package that teaches AI to generate Python-specific packages:
---
name: python-package-creator
description: Use when creating PRPM packages for Python best practices, frameworks, or libraries
---
# Python Package Creator
## When to Use
Triggers when user says:
- "Create a package for Python {topic}"
- "Generate Python {framework} rules"
- "Package my Python best practices"
## Package Structure Standards
### For Python Skills (Claude)
```markdown
---
name: {framework}-expert
description: Use when working with {framework} - covers common patterns, pitfalls, and best practices
---
# {Framework} Expert
## When to Use
- Starting a new {framework} project
- Debugging {framework} errors
- Optimizing {framework} performance
## Common Patterns
### Pattern 1: {Real Pattern Name}
**Use case**: [Specific scenario]
**Example**:
```python
# Actual working code
### Required Metadata
Always include:
- `tags`: ["python", "{framework}", "best-practices"]
- `format`: "claude" or "cursor"
- `subtype`: "skill" or "rule"
- Real code examples that actually work
### CSO Optimization
Include searchable keywords:
- Framework names: flask, django, fastapi
- Common errors: "ModuleNotFoundError", "circular import"
- Use cases: "async", "database migrations", "testing"
## Quality Checklist
Before generating, ensure:
- [ ] Real code examples (not pseudocode)
- [ ] Specific error messages mentioned
- [ ] Clear triggering conditions
- [ ] Proper frontmatter
- [ ] Relevant tags
## Example Output
```markdown
---
name: fastapi-expert
description: Use when building FastAPI applications or encountering async route errors
---
# FastAPI Expert
## When to Use
- Building REST APIs with FastAPI
- Debugging async route handlers
- Implementing dependency injection
- Working with Pydantic models
## Common Patterns
### Dependency Injection
**Use case**: Database connections, auth, logging
```python
from fastapi import Depends
from sqlalchemy.orm import Session
def get_db():
db = SessionLocal()
try:
yield db
finally:
db.close()
@app.get("/users/{user_id}")
async def get_user(user_id: int, db: Session = Depends(get_db)):
return db.query(User).filter(User.id == user_id).first()
get_db() automatically, handles cleanup
## Anti-Patterns to Avoid
❌ Generic examples without context
❌ Incomplete code snippets
❌ No explanation of when to use
❌ Missing error messages
❌ Vague descriptions
.claude/skills/python-package-creator/SKILL.md and publish:
Real-World Use Case: PRPM’s Own Development
PRPM uses meta packages to generate its internal packages. Here’s the workflow:
prpm install @prpm/creating-skills-skill
2. Create Package
User: "Create a skill for Stripe webhook integration"
AI: *references creating-skills-skill*
AI: *generates structured skill with:*
- Proper frontmatter
- CSO-optimized description
- Real webhook code examples
- Common error solutions
3. Publish Package
The generated package is now available in the PRPM registry for others to use.
Pattern 1: Template-Based Generation
Provide a complete template with placeholders:
## Template
```markdown
---
name: {{framework}}-{{feature}}
description: Use when {{use_case}}
tags: ["{{framework}}", "{{category}}", "best-practices"]
---
# {{Framework}} {{Feature}}
## When to Use
- {{trigger_condition_1}}
- {{trigger_condition_2}}
## Examples
{{real_code_example}}
### Pattern 2: Quality Enforcement
Require specific quality standards:
```markdown
## Quality Requirements
Before generating, verify:
1. All code examples are tested and working
2. Error messages are exact (not paraphrased)
3. Framework version is specified
4. At least 3 real-world use cases included
5. Links to official docs provided
Pattern 3: CSO Optimization
Teach AI to optimize for searchability:
## CSO Keywords
Include these in description:
- Exact error messages users will search for
- Framework-specific terms
- Common task descriptions
- Integration names
Example:
✅ "Use when getting 'Raw body not available' in Stripe webhooks"
❌ "Use for Stripe integration"
## Format Selection
**Cursor (.mdc files)**:
- Use frontmatter with `ruleType`
- Keep concise (Cursor prefers shorter rules)
- Focus on actionable patterns
**Claude (SKILL.md files)**:
- More detailed explanations allowed
- Include "When to Use" section
- Can be longer with examples
**Continue (.continuerc.json)**:
- JSON-based slash commands
- Focus on prompt engineering
- Include system messages
# Install meta-meta package
prpm install @prpm/meta-package-creator
# Now AI can create packages that teach how to create packages
User: "Create a meta package for Rust best practices"
AI: *generates rust-package-creator that teaches AI how to make Rust packages*
- Meta package teaches AI to create packages
- Those packages teach AI domain expertise
- Domain expertise helps users solve problems
- Users share new patterns back to packages
Before publishing, test that your meta package actually works:
prpm install @your-org/python-package-creator
2. Generate a Package
User: "Create a FastAPI authentication package"
AI: *uses python-package-creator guidance*
AI: *generates package*
3. Verify Output Quality
4. Iterate
If output is poor, improve the meta package’s guidance.
Best Practices
1. Provide Complete Templates
Show exactly what good packages look like, with all sections filled in.
2. Use Real Examples
Don’t use placeholder text. Show actual packages from the registry as examples.
3. Enforce Standards
Include checklists and requirements that AI must follow.
4. Optimize for Discovery
Teach AI to include CSO keywords that users will actually search for.
5. Version Your Guidance
As PRPM evolves, update meta packages to reflect new best practices.
Common Pitfalls
❌ Too Abstract
"Create a well-structured package with good examples"
✅ Specific and Actionable
"Include frontmatter with name, description, and tags array.
Add 3 code examples with comments.
Ensure description mentions exact error messages."
❌ Generic Examples
"Example: Some code here..."
✅ Real, Tested Code
"Example:
```python
from fastapi import FastAPI, HTTPException
app = FastAPI()
@app.get("/users/{user_id}")
async def get_user(user_id: int):
# This pattern avoids N+1 queries
user = await db.users.find_one({"id": user_id})
if not user:
raise HTTPException(status_code=404, detail="User not found")
return user
## Publishing Your Meta Package
```bash
# Create prpm.json
{
"name": "python-package-creator",
"description": "Meta package for creating high-quality Python PRPM packages",
"tags": ["meta", "python", "package-creation", "ai-prompts"],
"format": "claude",
"subtype": "skill",
"files": [".claude/skills/python-package-creator/SKILL.md"]
}
# Publish
prpm publish
