# Prompt Composition Template
Use a minimal, high-signal prompt structure. Keep only what the task needs.
## Template
```markdown
# Objective
- Do: [what to do]
- Don't: [what to avoid]
# Input
- Data: [files, text, or references]
- Assumptions: [constraints or environment]
# Output Format
- Format: [bullet list / JSON / table]
- Length: [max items / max tokens]
# Examples (1–3 representative)
- Example 1: [input → output]
- Note: Explain why this is a good example
# Validation Criteria
- Test: [test name or procedure]
- Expected: [exact pass condition]
- Fail: [what to do on failure]
# Escape Hatch
- If missing or uncertain, return: "Not found" (or specified fallback)
# Additional Context
- Only the minimum needed context
```
## Usage Guidelines
- **Objective**: Always start with clear Do/Don't boundaries
- **Examples**: 1-3 representative examples are more effective than lengthy explanations
- **Validation**: Define measurable success criteria upfront
- **Escape Hatch**: Prevent hallucination by providing explicit fallback behavior
- **description (REQUIRED)**: Every `.prompt.md` must have a `description` in YAML frontmatter — it appears in the VS Code prompt picker UI and helps both humans and AI identify the prompt's purpose
```yaml
---
description: One-line summary of what this prompt does
---
```
## Minimal Template (Quick Use)
```markdown
# Task
[One sentence objective]
# Input
[Data source]
# Output
[Format: JSON/Markdown/etc.]
[Max length if applicable]
# Example
Input: [x] → Output: [y]
```
