Standard Prompt Template
Title: Clear, searchable name
Purpose: 1—2 sentences (what it solves)
Prompt Body: Full text with placeholders
Inputs: What the user provides
Output Format: Expected structure (JSON, bullets, etc.)
Model & Settings: Recommended model, temperature, etc.
Examples: 1—2 input→output examples
Tags: #category #use-case
Owner: Who maintains this
Version: Current version
Last Updated: Date
Known Limitations: Where it fails
Example: Classify Sentiment
Title: Sentiment Classification
Purpose: Classify customer feedback as positive, negative, or neutral
Prompt Body: "Classify the following feedback as positive, negative, or neutral. Respond with only one word.\n\nFeedback: {feedback}"
Inputs: feedback (string)
Output Format: "positive" | "negative" | "neutral"
Model: GPT-4o or Claude
Examples: "Great service!" → "positive", "Broke after 1 day" → "negative"
Version: v1.2 (fixed sarcasm detection)
Owner: Customer Success team
Where to Store Documentation
- Git: Markdown files, version control
- Spreadsheet: Searchable, non-technical
- Wiki: Central knowledge base
- Prompt tool: Built-in documentation
Make Prompts Searchable
- Use consistent naming
- Tag heavily (#classification, #sentiment)
- Write clear descriptions
- Organize by use case or domain
Sources
- Markdown. Writing style guide
- GitHub. README best practices
- Notion. Template design
Common Mistakes
- Documentation out of sync with actual prompt
- No examples (confusing for new users)
- Poor naming (unmemorable, unsearchable)
- No metadata (can't find it later)
- Overly detailed (people skip it)
- No limitations section (hidden failures)