📝 Add prompt engineering guidelines fragments

Author: thibaultyou

fragments/prompt_engineering_compact_guidelines.md

@@ -0,0 +1,254 @@
+# Compact Guide to Prompt Engineering with AI Assistants
+
+## Introduction
+
+This guide provides strategies and best practices to effectively interact with AI assistants, enhancing their output accuracy and quality.
+
+## Table of Contents
+
+1. [Be Clear, Direct, and Detailed](#be-clear-direct-and-detailed)
+2. [Use Examples to Guide Behavior](#use-examples-to-guide-behavior)
+3. [Encourage Step-by-Step Thinking](#encourage-step-by-step-thinking)
+4. [Use XML Tags to Structure Prompts](#use-xml-tags-to-structure-prompts)
+5. [Assign a Role Using System Prompts](#assign-a-role-using-system-prompts)
+6. [Prefill Responses for Output Control](#prefill-responses-for-output-control)
+7. [Chain Prompts for Complex Tasks](#chain-prompts-for-complex-tasks)
+8. [Tips for Long Contexts](#tips-for-long-contexts)
+
+---
+
+## Be Clear, Direct, and Detailed
+
+Treat the AI assistant as a new employee who needs explicit instructions. Provide precise and detailed prompts to get better responses.
+
+- **Golden Rule:** If a colleague with minimal context finds your prompt confusing, the assistant likely will too.
+- **How to Be Clear:**
+  - Provide context (purpose, audience, workflow, goals).
+  - Be specific about what you want.
+  - Use numbered lists or bullet points for instructions.
+
+**Example: Anonymizing Customer Feedback**
+
+*Clear Prompt:*
+
+```markdown
+Your task is to anonymize customer feedback for our quarterly review.
+
+Instructions:
+1. Replace all customer names with "CUSTOMER_[ID]".
+2. Replace email addresses with "EMAIL_[ID]@example.com".
+3. Redact phone numbers as "PHONE_[ID]".
+4. Leave product names intact.
+5. If no PII is found, copy the message verbatim.
+6. Output only the processed messages, separated by "---".
+
+Data to process: {{FEEDBACK_DATA}}
+```
+
+---
+
+## Use Examples to Guide Behavior
+
+Including relevant examples helps the assistant understand the desired output format and style.
+
+- **Benefits:**
+  - Improves accuracy and consistency.
+  - Helps with complex tasks requiring specific formats.
+
+- **Crafting Examples:**
+  - Make them relevant and diverse.
+  - Wrap examples in `<example>` tags.
+
+**Example: Analyzing Customer Feedback**
+
+*Prompt with Example:*
+
+```markdown
+Our CS team needs to categorize feedback. Use categories: UI/UX, Performance, Feature Request, Integration, Pricing, Other. Rate sentiment and priority.
+
+<example>
+Input: The new dashboard is confusing and slow.
+Category: UI/UX, Performance
+Sentiment: Negative
+Priority: High
+</example>
+
+Now, analyze this feedback: {{FEEDBACK}}
+```
+
+---
+
+## Encourage Step-by-Step Thinking
+
+For complex tasks, instruct the assistant to think through the problem step-by-step.
+
+- **Benefits:**
+  - Increases accuracy and coherence.
+  - Helps with debugging by revealing the thought process.
+
+- **How to Prompt:**
+  - Include phrases like "Think step-by-step."
+  - Use structured prompts with `<thinking>` and `<answer>` tags.
+
+**Example: Financial Advice**
+
+*Prompt:*
+
+```markdown
+You're a financial advisor. A client has $10,000 to invest and needs it in 5 years for a house down payment. Option A: Volatile stock with 12% annual return. Option B: Bond with guaranteed 6% annual return. Which do you recommend? Think step-by-step.
+```
+
+*Assistant's Response:*
+
+```markdown
+<thinking>
+[Step-by-step analysis]
+</thinking>
+
+<answer>
+[Recommendation]
+</answer>
+```
+
+---
+
+## Use XML Tags to Structure Prompts
+
+XML tags help organize prompts with multiple components, improving clarity and reducing errors.
+
+- **Benefits:**
+  - Clearly separates instructions, context, and examples.
+  - Enhances parseability of outputs.
+
+- **Best Practices:**
+  - Be consistent with tag names.
+  - Nest tags appropriately.
+
+**Example: Generating Financial Reports**
+
+*Prompt:*
+
+```markdown
+You're a financial analyst at AcmeCorp.
+
+<data>{{SPREADSHEET_DATA}}</data>
+
+<instructions>
+1. Include sections: Revenue Growth, Profit Margins, Cash Flow.
+2. Highlight strengths and areas for improvement.
+</instructions>
+
+Use a concise and professional tone.
+
+<formatting_example>{{Q1_REPORT}}</formatting_example>
+```
+
+---
+
+## Assign a Role Using System Prompts
+
+Setting a role in the system prompt tailors the assistant's responses to fit a specific persona or expertise.
+
+- **Benefits:**
+  - Enhances accuracy in specialized tasks.
+  - Adjusts tone and focus.
+
+- **How to Assign a Role:**
+
+Include a role description in the system parameter of your API call.
+
+```python
+import anthropic
+
+client = anthropic.Anthropic()
+
+response = client.messages.create(
+    model="assistant-model",
+    max_tokens=2048,
+    system="You are the General Counsel of a Fortune 500 tech company.",
+    messages=[
+        {"role": "user", "content": "Analyze this contract..."}
+    ]
+)
+
+print(response.content)
+```
+
+---
+
+## Prefill Responses for Output Control
+
+By prefilling part of the assistant's response, you can guide the format and content of the output.
+
+- **Benefits:**
+  - Controls formatting (e.g., JSON, XML).
+  - Helps maintain character consistency.
+
+- **How to Prefill:**
+
+Include the initial text in the assistant's response field before generating the reply.
+
+**Example:**
+
+*Prefill:*
+
+```markdown
+Assistant: {
+```
+
+*Assistant's Completion:*
+
+```markdown
+"name": "Product Name",
+"price": "$49.99",
+"colors": ["black", "white"]
+}
+```
+
+---
+
+## Chain Prompts for Complex Tasks
+
+Break down complex tasks into smaller subtasks handled sequentially or in parallel.
+
+- **Benefits:**
+  - Improves accuracy and clarity.
+  - Easier to debug and refine.
+
+- **How to Chain Prompts:**
+  - Identify and separate subtasks.
+  - Use XML tags to pass outputs between prompts.
+  - Keep each subtask focused.
+
+**Example: Legal Contract Analysis**
+
+1. **Prompt 1:** Analyze the contract and output findings in `<risks>` tags.
+2. **Prompt 2:** Draft an email based on `<risks>`.
+3. **Prompt 3:** Review the email for tone and clarity.
+
+---
+
+## Tips for Long Contexts
+
+When working with large amounts of data, follow these guidelines.
+
+- **Place Long Data at the Top:** Include extensive documents before your query.
+- **Structure with XML Tags:** Organize multiple documents using tags for clarity.
+- **Ground Responses in Quotes:** Ask the assistant to reference specific parts of the data.
+
+**Example Structure:**
+
+```markdown
+<documents>
+  <document>
+    <source>Document 1</source>
+    <content>{{DOC1_CONTENT}}</content>
+  </document>
+  <document>
+    <source>Document 2</source>
+    <content>{{DOC2_CONTENT}}</content>
+  </document>
+</documents>
+
+[Your query or task]
+```