Anthropic revealed the anatomy of the perfect Claude prompt

Your Claude outputs will never look the same after this:

(taken from Anthropic’s Prompting 101)

1. Task context: Tells the model who it is and what it's supposed to do.

2. Tone context: Sets the style or personality of the response.

3. Background data, documents, and images: Provides the reference material the model should use to answer.

4. Detailed task description & rules: Spells out specific constraints and behaviors the model must follow.

5. Examples: Shows the model exactly what a good response looks like.

6. Conversation history: Gives the model the prior back-and-forth so it has context.

7. Immediate task description or request: The actual question or task the user is asking right now.

8. Thinking step by step / take a deep breath: Tells the model to reason carefully before answering.

9. Output formatting: Specifies the structure or format you want the answer in.

10. Prefilled response (if any): Starts the model's reply with a specific opening to steer the output.

––––––

→ Best Claude guides: docs.google.com/documen…

♻️ Restack this if you learned something new