Prompt Library Curator

Overview

Organize, version, and share effective prompts across the organization. Maintain quality standards, track usage, and enable prompt reuse.

Core principle: Good prompts are organizational assets. Curate them like code—versioned, documented, and shared.

When to Use

•Building shared prompt repository
•Standardizing prompts across teams
•Documenting successful prompt patterns
•Onboarding new prompt engineers

Output Format

yaml

prompt_library:
  version: "[Library version]"
  last_updated: "[YYYY-MM-DD]"
  
  prompts:
    - id: "[PROMPT-XXX]"
      name: "[Descriptive name]"
      version: "[X.Y.Z]"
      
      metadata:
        author: "[Creator]"
        created: "[YYYY-MM-DD]"
        updated: "[YYYY-MM-DD]"
        category: "[Classification | Extraction | Generation | etc.]"
        tags: ["[tag1]", "[tag2]"]
        status: "[Draft | Active | Deprecated]"
      
      prompt:
        system: |
          [System prompt content]
        
        user_template: |
          [User prompt template with {{variables}}]
        
        variables:
          - name: "{{variable_name}}"
            description: "[What it represents]"
            type: "[string | list | json]"
            required: [true | false]
            example: "[Example value]"
      
      model_compatibility:
        tested_models: ["[Model 1]", "[Model 2]"]
        recommended_model: "[Best performing model]"
        parameters:
          temperature: "[Value]"
          max_tokens: "[Value]"
          other: "[Other params]"
      
      performance:
        quality_rating: "[1-5]"
        reliability: "[Consistent | Variable | Unstable]"
        latency: "[Fast | Medium | Slow]"
        cost_tier: "[Low | Medium | High]"
      
      usage:
        use_cases: ["[Use case 1]", "[Use case 2]"]
        anti_patterns: ["[When NOT to use]"]
        example_inputs: ["[Example]"]
        example_outputs: ["[Example]"]
      
      evaluation:
        criteria: ["[How to judge quality]"]
        test_cases:
          - input: "[Test input]"
            expected: "[Expected output pattern]"
      
      related_prompts: ["[PROMPT-XXX]"]
      
      changelog:
        - version: "[X.Y.Z]"
          date: "[YYYY-MM-DD]"
          changes: "[What changed]"
  
  categories:
    - name: "[Category name]"
      description: "[What prompts in this category do]"
      prompt_count: "[N]"
  
  guidelines:
    quality_standards: "[Link or embedded standards]"
    contribution_process: "[How to add new prompts]"
    review_process: "[How prompts are reviewed]"

Prompt Categories

Category	Purpose	Examples
Classification	Categorize inputs	Sentiment, intent, topic
Extraction	Pull structured data	Entity extraction, parsing
Generation	Create content	Summaries, drafts, responses
Transformation	Convert formats	Translation, reformatting
Analysis	Evaluate or assess	Quality scoring, comparison
Reasoning	Multi-step thinking	Problem solving, planning

Prompt Quality Standards

Quality Rating (1-5)

Rating	Criteria
5	Production-ready, tested extensively, documented
4	Reliable, good coverage, minor edge cases
3	Works for main cases, needs refinement
2	Experimental, inconsistent results
1	Draft, untested

Quality Checklist

yaml

quality_checklist:
  clarity:
    - "Instructions are unambiguous"
    - "Expected output format is clear"
    - "Edge cases are addressed"
  
  robustness:
    - "Tested with diverse inputs"
    - "Handles malformed input gracefully"
    - "Consistent outputs across runs"
  
  documentation:
    - "Purpose clearly stated"
    - "Variables documented"
    - "Examples provided"
    - "Anti-patterns noted"

Versioning Strategy

Semantic Versioning for Prompts

code

v[MAJOR].[MINOR].[PATCH]

MAJOR: Significant behavior change, output format change
MINOR: New capabilities, improved handling
PATCH: Typo fixes, minor wording improvements

When to Version

Change	Version Impact
Output format change	MAJOR
New variable added	MINOR
Better edge case handling	MINOR
Typo fix	PATCH
Wording clarification	PATCH

Prompt Template Best Practices

Structure

yaml

prompt_structure:
  role_definition: "Who the AI is acting as"
  context: "Background information needed"
  task: "What to do"
  constraints: "Rules and limitations"
  output_format: "How to structure the response"
  examples: "Few-shot examples if needed"

Example Template

markdown

# System Prompt Template

You are a {{role}} helping with {{task_domain}}.

## Your Task
{{task_description}}

## Guidelines
- {{guideline_1}}
- {{guideline_2}}

## Output Format
Respond with a JSON object:
```json
{
  "field1": "description",
  "field2": "description"
}

Examples

Input: {{example_input}} Output: {{example_output}}

code


## Usage Tracking

```yaml
usage_metrics:
  prompt_id: "[PROMPT-XXX]"
  period: "[Month/Quarter]"
  
  metrics:
    invocations: "[Count]"
    unique_users: "[Count]"
    teams_using: ["[Team 1]", "[Team 2]"]
    
  quality_feedback:
    thumbs_up: "[Count]"
    thumbs_down: "[Count]"
    comments: ["[Feedback]"]
    
  issues_reported: "[Count]"
  improvements_suggested: ["[Suggestion]"]

Contribution Process

yaml

contribution_workflow:
  1_draft:
    - "Create prompt following template"
    - "Document variables and examples"
    - "Self-test with diverse inputs"
  
  2_review:
    - "Submit for peer review"
    - "Address feedback"
    - "Test with reviewer's inputs"
  
  3_approval:
    - "Quality rating assigned"
    - "Category and tags confirmed"
    - "Added to library with Draft status"
  
  4_promotion:
    - "Pilot with limited users"
    - "Gather feedback"
    - "Promote to Active if successful"

Common Anti-Patterns

Anti-Pattern	Problem	Fix
Vague instructions	Inconsistent outputs	Be specific about expectations
No output format	Unparseable responses	Specify exact format needed
Missing examples	Model guesses intent	Add few-shot examples
Too many constraints	Model gets confused	Prioritize key constraints
Outdated prompts	Uses deprecated patterns	Regular review and updates

Checklist

• Prompt follows template structure
• All variables documented
• Examples provided (input and output)
• Model compatibility tested
• Quality rating assigned
• Category and tags applied
• Version tracked
• Usage guidelines documented