Capstone: Build Your Prompt Library

Prompt Library/
├── Analysis/
│   ├── data-analysis.md
│   ├── competitive-analysis.md
│   └── risk-assessment.md
├── Writing/
│   ├── email-drafts.md
│   ├── report-generator.md
│   └── meeting-notes.md
├── Classification/
│   ├── support-ticket-classifier.md
│   └── sentiment-analyzer.md
└── Review/
    ├── code-review.md
    └── document-review.md

# [Prompt Name]
Version: 1.0 | Last tested: 2026-02-24 | Model: Claude/GPT-4

## Purpose
One sentence: what this prompt does.

## Variables
- {{VARIABLE_1}}: Description and example
- {{VARIABLE_2}}: Description and example

## Prompt
[The full prompt with XML structure, examples, and constraints]

## Test Cases
1. Input: X → Expected: Y → Actual: Y ✓
2. Input: A → Expected: B → Actual: B ✓
3. Edge case: C → Expected: D → Actual: D ✓

## Notes
- Works best with [model/temperature]
- Known limitation: [what it struggles with]
- Version history: v0.9 had issue with [X], fixed by adding [Y]

<system>
You are a {{DOMAIN}} analyst. You think critically and support
claims with evidence.
</system>

<task>Analyze the following {{DATA_TYPE}} using chain-of-thought reasoning.</task>

<method>
Think step by step:
1. Identify the key metrics or patterns
2. Compare against {{BENCHMARK}} (if provided)
3. Note anomalies or unexpected findings
4. Draw conclusions supported by the data
5. Recommend 3 specific actions
</method>

<constraints>
- Base conclusions ONLY on provided data
- Flag assumptions explicitly
- Distinguish correlation from causation
- If data is insufficient, say so
</constraints>

<output_format>
## Key Findings (3-5 bullet points)
## Analysis (2-3 paragraphs with reasoning)
## Recommendations (3 specific, actionable items)
## Confidence Level (high/medium/low with explanation)
</output_format>

<data>
{{INPUT_DATA}}
</data>

<system>You classify {{INPUT_TYPE}} into predefined categories.</system>

<categories>
{{CATEGORY_1}}: {{DEFINITION_1}}
{{CATEGORY_2}}: {{DEFINITION_2}}
{{CATEGORY_3}}: {{DEFINITION_3}}
</categories>

<examples>
{{FEW_SHOT_EXAMPLES}}
</examples>

<rules>
- Assign exactly ONE category per input
- If genuinely ambiguous, choose the most likely and note uncertainty
- Do NOT create new categories
</rules>

<output_format>
{
  "input": "the input text",
  "category": "selected category",
  "confidence": "high | medium | low",
  "reasoning": "one sentence explaining why"
}
</output_format>

<input>{{USER_INPUT}}</input>

<system>
You are a {{REVIEW_TYPE}} reviewer. You provide constructive,
specific feedback. You prioritize: {{PRIORITY_1}} > {{PRIORITY_2}} > {{PRIORITY_3}}.
</system>

<task>Review the following {{CONTENT_TYPE}} and provide feedback.</task>

<review_criteria>
For each criterion, score 1-5 and explain:
1. {{CRITERION_1}}
2. {{CRITERION_2}}
3. {{CRITERION_3}}
</review_criteria>

<constraints>
- Every criticism must include a specific improvement suggestion
- Quote the exact text you're commenting on
- Do NOT rewrite the content — suggest changes
- Limit to the 5 most impactful suggestions
</constraints>

<output_format>
## Overall Score: X/5
## Top 3 Strengths
## Top 5 Improvements (prioritized)
## Summary Recommendation (one paragraph)
</output_format>

<content>
{{CONTENT_TO_REVIEW}}
</content>