03. Prompt Anatomy

flowchart TD
    subgraph BAD["Unstructured Prompt"]
        B1["'Write code for a task manager app'"] --> B2["LLM guesses:\n-Frontend? Backend? API?\n-Which language? Framework?\n-What features?"]
        B2 --> B3["❌ Random, incomplete output"]
    end

    subgraph GOOD["Structured Prompt"]
        G1["Instruction: Build a REST API\nTech: Node.js + Express\nFeatures: CRUD for tasks\nFormat: Return code with explanations"] --> G2["LLM follows spec"]
        G2 --> G3["✅ Complete, correct output"]
    end

    style BAD fill:#ef4444,color:#fff
    style GOOD fill:#22c55e,color:#fff

Ticket Title: Add user authentication
Description: Users should be able to log in with email/password
Acceptance Criteria:
  - Login form with email + password fields
  - Validation on both fields
  - Error messages for invalid credentials
  - Redirect to dashboard on success
Tech Notes: Use JWT tokens, store in httpOnly cookies

flowchart TD
    PROMPT["Complete Prompt"] --> I["1. Instruction\nWhat to do"]
    PROMPT --> C["2. Context\nBackground info"]
    PROMPT --> D["3. Input Data\nThe data to process"]
    PROMPT --> CON["4. Constraints\nRules & limits"]
    PROMPT --> F["5. Output Format\nHow to respond"]
    PROMPT --> E["6. Examples\nShow, don't just tell"]
    PROMPT --> P["7. Persona\nWho the AI should be"]

    style I fill:#3b82f6,color:#fff
    style C fill:#f59e0b,color:#fff
    style D fill:#22c55e,color:#fff
    style CON fill:#ef4444,color:#fff
    style F fill:#8b5cf6,color:#fff
    style E fill:#ec4899,color:#fff
    style P fill:#14b8a6,color:#fff

flowchart LR
    subgraph WITHOUT["Without Context"]
        WA["User: 'Is this query efficient?'"] --> WB["LLM: 'It depends...'"]
        WB --> WC["❌ Generic, unhelpful answer"]
    end

    subgraph WITH["With Context"]
        WD["User: 'Is this query efficient?'\nContext: 'MySQL 8.0, 10M rows in 'orders' table, \nindex on (customer_id, order_date), query:\nSELECT * FROM orders WHERE customer_id = 5'"] --> WE["LLM: 'Yes, the index on customer_id\nmakes this efficient. However, SELECT *\ncould be optimized...'"]
        WE --> WF["✅ Specific, actionable advice"]
    end

    style WITHOUT fill:#ef4444,color:#fff
    style WITH fill:#22c55e,color:#fff

Instruction: Summarize this article
Context: The summary should be 3 sentences for a busy executive
Input Data: [The article text goes here]

❌ Mixed: "Summarize this article about AI written by John Smith on
    March 15th where he talks about transformers and attention
    mechanisms and compares them to CNNs..."

✅ Structured:
    Task: Summarize the following article
    Article: [paste article text]
    Requirements: 3 sentences, focus on key findings, avoid technical jargon

flowchart TD
    subgraph NOCON["Without Constraints"]
        N1["'Explain Docker'"] --> N2["5 paragraphs on container history"]
        N2 --> N3["3 paragraphs on installation"]
        N3 --> N4["Endless, unstructured output"]
    end

    subgraph WITHCON["With Constraints"]
        W1["'Explain Docker in 3 bullet points.\nAssume I know what a VM is.\nFocus on why Docker, not how to install.'"] --> W2["1. Lightweight vs VMs\n2. Consistent environments\n3. Easy scaling"]
        W2 --> W3["Concise, targeted output"]
    end

    style NOCON fill:#ef4444,color:#fff
    style WITHCON fill:#22c55e,color:#fff

❌ "List the top 5 features of React 19"
   → Might return: paragraph, bullet list, numbered list, or random format

✅ "List the top 5 features of React 19 in this exact format:
   | # | Feature | Description | Impact |
   |---|---------|-------------|--------|
   | 1 | [name]  | [1 sentence] | [high/med] |"
   → Returns: the exact table you requested

flowchart LR
    subgraph ZERO["Zero-Shot (No Example)"]
        Z1["'Classify this email'"] --> Z2["❌ May guess wrong format"]
    end

    subgraph ONE["One-Shot (1 Example)"]
        O1["'Classify this email'\nExample: 'Meeting at 3pm' → 'Meeting'"] --> O2["✅ Follows the pattern"]
    end

    subgraph FEW["Few-Shot (3+ Examples)"]
        F1["'Classify this email'\nEx1: 'Meeting at 3' → 'Meeting'\nEx2: 'Bug in login' → 'Bug'\nEx3: 'Deploy to prod' → 'Deploy'"] --> F2["✅ Very accurate pattern matching"]
    end

    style ZERO fill:#f59e0b,color:#fff
    style ONE fill:#3b82f6,color:#fff
    style FEW fill:#22c55e,color:#fff

❌ "Explain microservices"
   → Could be: academic, tutorial, pros-and-cons, or sales pitch

✅ "You are a Staff Engineer at Netflix with 15 years of experience
    building microservices. Explain microservices to a junior developer
    on your team who needs to understand the trade-offs."
   → Response will be: practical, trade-off aware, from-experience

Persona: [Who the AI should be]
Context: [Background information]
Task: [Specific instruction]
Constraints:
  - [Rule 1]
  - [Rule 2]
Input Data:
[The data to process]
Output Format:
[Desired format with example if possible]

Persona: You are a Senior DevOps Engineer
Context: We're migrating a Node.js app from Heroku to AWS ECS
Task: Create a Dockerfile for the app
Constraints:
  - Use Node.js 20 Alpine as base
  - Multi-stage build for smaller image size
  - Include healthcheck
  - Don't expose ports (handled by ECS)
  - Optimize for production, not development
Output Format:
  Provide the Dockerfile in a code block, followed by a brief
  explanation of each stage and why specific choices were made.

[
  {"role": "system", "content": "You are a helpful coding assistant."},
  {"role": "user", "content": "Write a Python function to..."}
]

[
  {"role": "user", "content": "I'll provide context first, then my question."},
  {"role": "assistant", "content": "I understand. Go ahead."},
  {"role": "user", "content": "Context: [data]. Question: [question]"}
]