Créer une bibliothèque de prompts réutilisables

prompts/
├── README.md                 # Library documentation
├── _templates/               # Reusable base templates
│   ├── base-article.md
│   ├── base-email.md
│   └── base-analysis.md
├── marketing/
│   ├── prospecting-email.md
│   ├── followup-email.md
│   ├── landing-page.md
│   └── social-media/
│       ├── linkedin-post.md
│       └── twitter-thread.md
├── content/
│   ├── blog-seo.md
│   ├── newsletter.md
│   └── case-study.md
├── code/
│   ├── code-review.md
│   ├── refactoring.md
│   └── api-design.md
├── analysis/
│   ├── data-analysis.md
│   ├── competitor-analysis.md
│   └── market-research.md
└── system-prompts/
    ├── support-agent.md
    ├── seo-writer.md
    └── code-assistant.md

# B2B SaaS Prospecting Email

## Metadata
- **ID**: MKT-001
- **Version**: 2.3
- **Category**: Marketing > Email
- **Author**: Nicolas
- **Last Modified**: 2025-02-20
- **Tested Models**: Claude 3.5 Sonnet, GPT-4
- **Quality Score**: 8.5/10
- **Tags**: email, prospecting, B2B, SaaS

## Variables
| Variable | Description | Example |
|----------|-------------|---------|
| [[PROSPECT_NAME]] | Prospect's name | John Doe |
| [[COMPANY]] | Prospect's company | TechCorp |
| [[POSITION]] | Prospect's position | CTO |
| [[PAIN_POINT]] | Identified problem | Infrastructure costs |
| [[PRODUCT]] | Our product | CloudOptim |
| [[HOOK]] | Personalized hook | Your LinkedIn post about... |

## System Prompt
You are a senior B2B SaaS salesperson. You write prospecting emails that achieve >30% open rates and >10% response rates.

## Prompt
Write a prospecting email for [[PROSPECT_NAME]], [[POSITION]] at [[COMPANY]].

Context:
- Identified pain point: [[PAIN_POINT]]
- Our solution: [[PRODUCT]]
- Hook: [[HOOK]]

Constraints:
- Max 5 body lines
- Subject: max 6 words, curiosity without clickbait
- No "I'm reaching out", no "don't hesitate"
- CTA: one simple question
- Tone: professional but human

## Expected Output Example
[A concrete example of the ideal result]

## Version History
- v2.3 (2025-02-20): Added HOOK variable for personalization
- v2.2 (2025-02-10): Reduced to 5 lines (better response rate)
- v2.1 (2025-01-28): Added anti-cliché constraint
- v2.0 (2025-01-15): Complete redesign, AIDA structure
- v1.0 (2024-12-01): Initial version

# Template: SEO Blog Article

You are a senior SEO web writer.

Write an article for [[COMPANY]]'s blog.

**Topic**: [[TOPIC]]
**Main keyword**: [[MAIN_KEYWORD]]
**Secondary keywords**: {{SECONDARY_KEYWORDS[]}}
**Audience**: [[AUDIENCE]]
**Length**: {{WORD_COUNT:1500}} words
**Tone**: {{TONE:professional|conversational|expert}}

Structure:
- Optimized H1 title (max 60 characters)
- Introduction with reader's problem (150 words)
{{#IF INCLUDE_FAQ}}
- FAQ (5 schema.org friendly questions)
{{/IF}}
- Meta title (≤60 characters)
- Meta description (150-160 characters)

import re

def fill_prompt(template: str, variables: dict) -> str:
    """Replace variables in a prompt template."""
    result = template
    for key, value in variables.items():
        if isinstance(value, list):
            value = ", ".join(value)
        result = result.replace(f"{{{[[key]]}}}", str(value))

    # Handle default values {{VAR:default}}
    result = re.sub(
        r'\{\{(\w+):([^}]+)\}\}',
        lambda m: variables.get(m.group(1), m.group(2)),
        result
    )
    return result

# Usage
template = open("prompts/content/blog-seo.md").read()
prompt = fill_prompt(template, {
    "COMPANY": "TechFlow",
    "TOPIC": "The 10 SaaS Pricing Mistakes",
    "MAIN_KEYWORD": "SaaS pricing",
    "SECONDARY_KEYWORDS": ["pricing", "freemium model", "pricing strategy"],
    "AUDIENCE": "SaaS startup founders",
    "TONE": "conversational"
})

marketing/      → Everything related to acquisition and conversion
content/        → Content writing (blog, social, newsletter)
code/           → Development, review, architecture
analysis/       → Data analysis, market, competition
operations/     → Process, documentation, project management
sales/          → Prospecting, qualification, closing
hr/             → Recruitment, onboarding, evaluation
system-prompts/ → System prompts for agents/chatbots

## Compatibility
- Claude 3.5 Sonnet: ✅ Tested, score 9/10
- GPT-4 Turbo: ✅ Tested, score 7/10 (tends to be too long)
- Llama 3 70B: ⚠️ Works but loses constraints
- Mistral Large: ✅ Tested, score 8/10

# Initialize repo
git init prompts-library
cd prompts-library

# Initial structure
mkdir -p {marketing,content,code,analysis,system-prompts,_templates}
touch README.md

# Initial commit
git add .
git commit -m "init: prompt library structure"

# New prompt version
git commit -m "feat(marketing): prospecting-email v2.3 - added HOOK variable"

# Tag for stable versions
git tag -a v1.0 -m "Version 1.0 - 25 tested and validated prompts"

feat(category): description     → New prompt or feature
fix(category): description      → Prompt correction
perf(category): description     → Performance improvement
test(category): description     → Added tests/examples
docs(category): description     → Documentation

                    ┌──────────┐
                    │  Draft   │
                    │ (draft)  │
                    └────┬─────┘
                         │ Test with 10 cases
                         ▼
                    ┌──────────┐
                    │   Beta   │
                    │ (v0.x)   │
                    └────┬─────┘
                         │ Peer validation
                         ▼
                    ┌──────────┐
                    │ Stable   │
                    │ (v1.0)   │
                    └────┬─────┘
                         │ Real-world feedback
                         ▼
                    ┌──────────┐
                    │ Iteration│
                    │ (v1.x)   │
                    └──────────┘

import random
import json
from datetime import datetime

class PromptABTest:
    def __init__(self, name: str, variants: dict):
        self.name = name
        self.variants = variants  # {"A": prompt_a, "B": prompt_b}
        self.results = {"A": [], "B": []}

    def get_variant(self) -> tuple:
        """Return a random variant."""
        variant = random.choice(["A", "B"])
        return variant, self.variants[variant]

    def log_result(self, variant: str, score: int, notes: str = ""):
        """Record a test result."""
        self.results[variant].append({
            "score": score,
            "notes": notes,
            "timestamp": datetime.now().isoformat()
        })

    def get_winner(self) -> str:
        """Determine the best variant."""
        avg_a = sum(r["score"] for r in self.results["A"]) / len(self.results["A"]) if self.results["A"] else 0
        avg_b = sum(r["score"] for r in self.results["B"]) / len(self.results["B"]) if self.results["B"] else 0
        return "A" if avg_a >= avg_b else "B", avg_a, avg_b

# _templates/base-article.md (parent template)

You are an expert writer for [[PUBLICATION]].

Style:
- Short sentences (max 20 words)
- Active voice
- Concrete examples
- No unexplained jargon

Standard structure:
- Optimized title
- Introduction (hook + promise)
- Body structured in H2/H3
- Conclusion with CTA

[[SPECIFIC_INSTRUCTIONS]]

# content/blog-tech.md (inherits from base-article)

{{INHERITS_FROM: _templates/base-article.md}}

PUBLICATION: TechFlow Blog
SPECIFIC_INSTRUCTIONS:
- Include code blocks when relevant
- Add "In summary" section with 3 bullet points
- Target junior-mid developer
- Link to official documentation