# The prompt pack — Claude Guide

Every prompt in the guide (catalogue, frameworks, Cowork, GDPR) in a single text file to keep at hand.
https://guide-claude.netlify.app/en/

## Module 03 — Prompt engineering — talking to Claude to get excellence
https://guide-claude.netlify.app/en/m/prompting/

```
# The complete prompt, ready to copy
Act as my chief of staff. This report is for my executive committee.
Task: summarize the decisions to make in 5 bullets, then list 3 risks
with their estimated financial impact. Format: 200 words maximum.
Guardrail: if a numerical data point is missing, write [to confirm] — don't make anything up.
```

```
Explain [topic] as if I had 15 minutes and no technical background.
What are the 3 points most people get wrong about [topic]?
Summarize this document in one sentence, then in a paragraph, then in 5 bullets.
```

```
Play devil's advocate on this decision: [decision]. Where are my blind spots?
Compare [option A] and [option B] in a table: cost, risk, timeline, reversibility.
If you were in my shoes, what would you do — and what could make you change your mind?
```

```
Rewrite this text for a [executive committee / client / general] audience, tone [direct / warm].
Give me 3 versions of this hook, from the most understated to the most striking.
Make this paragraph 30% shorter without losing any information.
```

```
Draft a polite but firm reply to this email: [paste the email].
Turn these notes into a courteous follow-up email that still gets a reply.
Help me say no to [request] without straining the relationship.
```

```
Turn this brief into an action plan with owners and deadlines.
From these meeting notes, extract only the decisions and the action items.
Prepare 5 tough questions my Board might ask about [topic].
```

## Module 03b — Decision frameworks — think better, faster
https://guide-claude.netlify.app/en/m/frameworks/

```
Run a SWOT analysis of [topic], taking into account [competitors / regulation / market]. End with the 3 priorities that follow from it.
```

```
Here are my tasks: [list]. Sort them into urgent/important (do, schedule, delegate, eliminate) and propose a plan for the week.
```

```
Analyze our industry [specify] with Porter's 5 forces (rivalry, new entrants, substitutes, buyer power, supplier power), then derive 3 levers for advantage.
```

```
Break down the problem [topic] into MECE categories (mutually exclusive, collectively exhaustive) and propose an action plan by category.
```

```
Our project [describe] will launch in [timeframe]. Assume it has failed: list the 8 most likely causes and one preventive fix for each.
```

```
Recurring problem: [describe]. Apply the 5 Whys method to identify the root cause, then propose a durable countermeasure.
```

```
Compare [options] against [criteria: cost, timeline, risk, impact] with weightings. Give a scored table and a justified recommendation.
```

```
Should we [decision]? Compare costs and benefits over 3 years vs. doing nothing, and recommend the best choice with the key assumptions.
```

```
Our offering [describe] is plateauing in [market]. Use the Ansoff matrix (penetration, product, market, diversification) and prioritize the most realistic axis.
```

```
We're expanding into [market]. Build 3 scenarios (optimistic, base case, pessimistic) over 3 years and a strategy that holds up in all three.
```

```
Our [line-item] cost is too high. Break it down into fundamental building blocks and rebuild an optimal approach from scratch.
```

```
Analyze the decision [topic] with the 6 hats: facts, risks, benefits, emotions, creativity, process. End with an actionable synthesis.
```

## Module 03 ter — Prompt or loop? — the shift
https://guide-claude.netlify.app/en/m/loops/

```
# A loop you can adapt — Monday morning competitive monitoring
TRIGGER: every Monday at 8 a.m.
TASK: report what changed at [competitor A], [competitor B], [competitor C]
      (website, pricing page, LinkedIn, press) since last Monday.
STOP CRITERION: every item kept has a clickable source, a date later than
      the previous Monday, and fits in one sentence.
      If nothing is new, write "nothing to report" — do not pad.
OUTPUT: prepend to monitoring.md, without overwriting the history.
GUARDRAIL: publish nothing, send nothing. I review before anything goes out.
```

## Module 05 — Cowork — your digital colleague
https://guide-claude.netlify.app/en/m/cowork/

```
You're going to become my colleague in Cowork. Before producing anything,
ask me one question at a time to get to know me well: who I am, my role,
my company, my clients, how I write, and the "too much AI" phrasings
I want to avoid. When you have enough information, create in this folder a
"my profile" sheet that will serve as your memory, then confirm to me that it's saved.
```

## Module 08 — Artifacts — build without coding
https://guide-claude.netlify.app/en/m/artifacts/

```
Create a React artifact (interactive application) named "Magic Prompteur": a tool that takes a raw prompt and rewrites it in the style of the world's best prompt engineer, drawing on a real AI embedded in the artifact (a call to the Anthropic API from the artifact via fetch, with no API key to pass).

HOW IT WORKS
- The user pastes their raw prompt into a text area.
- An "Expert settings" panel (collapsible) offers: Role/persona (free field), Tone (menu: Neutral, Professional, Punchy, Educational, Creative, Direct, Warm, Formal), Output format (Model's choice, Text, Bullet list, Numbered steps, Table, JSON, Markdown), Target model (Claude, ChatGPT/GPT, Gemini, Mistral, Midjourney/image, Generic), Level of detail (Concise, Balanced, Detailed, Ultra-detailed). All optional.
- An "Optimize my prompt" button sends it all to Claude.

OUTPUT: 3 optimized variants, each with a title, a one-sentence note explaining its angle, and a "Copy" button:
1. Precise & structured (clear constraints, explicit format)
2. Context-rich (strong role, examples, guardrails)
3. Chain of reasoning (step-by-step thinking)

API CALL WITHIN THE ARTIFACT: POST to https://api.anthropic.com/v1/messages, model claude-sonnet-5, max_tokens 8000. The system prompt makes Claude an elite prompt engineer who masters role/persona, context, constraints, format, few-shot, success criteria, and step-by-step reasoning. The rewritten prompts are in English, ready to paste, tailored to the target model.

IMPORTANT — ROBUSTNESS: do NOT use JSON for the output (the quotes, braces, and line breaks of the generated prompts break it). Use a delimiter format: each variant begins with "===VARIANTE===" with the fields "ID:", "TITRE:", "NOTE:" then "---PROMPT---" followed by the prompt text (which may contain any character). Parse by splitting on these delimiters, and display even partial variants if the response is truncated.

DESIGN: dark, premium interface (slate-950 background), violet/fuchsia/cyan gradients, lucide-react icons, subtle animations, fully responsive on mobile and desktop, in English. Each variant has its accent color (cyan/violet/amber) and its icon (target/layers/brain).

TECHNICAL CONSTRAINTS OF THE ARTIFACT: components defined outside the main component (otherwise fields lose focus on every keystroke). No localStorage/sessionStorage (not supported in artifacts) — state managed in memory via useState. Clean error handling with a user message. Copy button with execCommand fallback.
```

## Module 14 — 10 ways your data can leak — without any hacking at all
https://guide-claude.netlify.app/en/m/security/

```
# Sample prompt — analysis of a pseudonymized grid
Here is a pseudonymized compensation grid (S001 to S045),
with no names or identifiers. Analyze the gaps by level and by gender,
flag any anomaly beyond ±10% of the level's median,
and propose a compliance plan in 3 actions.
Do not attempt to identify the individuals.
```