← Back to Module 9
🏋️Module 9 — Exercise

Advanced Prompting Strategies

Decision frameworks for choosing the right mode (Ask, Edit, Agent) and the right model for every task.

Exercise: Advanced Prompting Strategies

Practice choosing the right mode, the right model, and writing effective prompts.

Task 1 — Mode Selection Drill

For each scenario below, decide which mode (Ask, Plan, or Agent) you would use and why. Write your answers down, then check the suggested answers at the bottom.

  1. You want to know what the useContext hook does in React.
  2. You need to change the page title in layout.tsx from "My App" to "Solar Academy".
  3. You want to add a complete new "Settings" page with a form, validation, and API integration.
  4. You see an error in the terminal and want to understand what it means.
  5. You want to plan a migration from REST to GraphQL before making any changes.
  6. You need to refactor the authentication flow across 5 files.
Suggested answers
  1. Ask — purely informational, no file changes needed
  2. Agent — straightforward file change, agent handles it quickly
  3. Agent — new page creation involves multiple files and possibly new dependencies
  4. Ask — you want an explanation, not a code change
  5. Plan — you want a step-by-step roadmap to review before committing to the migration
  6. Agent — multi-file refactoring requires reading and editing across the project (consider using Plan first to review the approach)

Task 2 — Model Switching Experiment

Perform the same task with two different models and compare the results:

  1. Open Copilot Chat in Ask mode
  2. Select GPT-4o as the model
  3. Type: "Explain 3 different approaches to implement dark mode in a Next.js app using Tailwind CSS. Compare the trade-offs."
  4. Read and note the response quality
  5. Switch to Claude Sonnet (use the model dropdown)
  6. Send the exact same prompt
  7. Compare:
    • Which response was more detailed?
    • Which gave better trade-off analysis?
    • Which was faster?

Takeaway: Notice how different models have different "thinking styles." Neither is wrong — they emphasize different things.

Task 3 — Prompt Refinement Chain

Practice the iterate, don't rewrite technique:

  1. Switch to Agent mode with Claude Sonnet
  2. Send: "Create a simple card component that shows a user's name and email"
  3. Once generated, send a follow-up: "Add an avatar image to the left side of the card"
  4. Then: "Make it responsive — stack vertically on mobile, horizontal on desktop"
  5. Then: "Add a hover effect that slightly elevates the card with a shadow"

Takeaway: Each message builds on the previous result. You don't need to describe the entire component from scratch in one mega-prompt.

Task 4 — Context → Task → Constraints Pattern

Practice structuring prompts with the three-part pattern:

  1. Open any file in your project
  2. In Copilot Chat (Ask mode), write a prompt using this structure:
Context: [describe the file/project/feature]
Task: [what you want done]
Constraints: [limitations, style rules, do-nots]

Example:

"This is a Next.js page that lists products from an API. Add pagination with Previous/Next buttons below the product grid. Use the existing Button component from @/components. Don't add any external pagination library."

  1. Send it and review the diff
  2. Try the same request but without context and constraints — notice how the output quality changes

Task 5 — The "Explain Then Do" Pattern

Practice the two-step workflow for a critical change:

  1. Switch to Ask mode
  2. Send: "I want to add form validation to a user registration form with fields: name (required, min 2 chars), email (required, valid format), and password (required, min 8 chars, must contain a number). What approach would you recommend?"
  3. Read the response carefully — does the approach make sense?
  4. If you agree, switch to Agent mode
  5. Send: "Go ahead and implement the registration form with the validation approach you just described"
  6. Review the generated code

Takeaway: The Ask step lets you validate the AI's approach before it writes code. This catches bad assumptions early.

Task 6 — Anti-Pattern Identification

Read each prompt below and identify what's wrong. Rewrite it as a better prompt.

Prompt A:

"Fix my app"

Prompt B:

"I need you to take the entire authentication system which currently uses JWT tokens stored in localStorage and migrate it to use HTTP-only cookies with a refresh token rotation strategy and also add CSRF protection and while you're at it update all the API routes to validate the new cookie-based auth and also add rate limiting to the auth endpoints and update the tests."

Prompt C:

"Add a button" (with no file open, no context attached)

Better versions

A rewritten: "The login form submits but the page doesn't redirect to the dashboard. The console shows a 401 error. #terminal What's causing this?"

B rewritten — broken into phases:

  • Phase 1: "Migrate the auth tokens from localStorage to HTTP-only cookies. Here's the current auth flow: #file:auth.ts"
  • Phase 2: "Now add refresh token rotation to the cookie-based auth"
  • Phase 3: "Add CSRF protection to all POST/PUT/DELETE routes"
  • Phase 4: "Add rate limiting to the auth endpoints"

C rewritten: "Add a 'Save Draft' button below the form in #file:EditPage.tsx. Use the Solar OutlinedButton component. On click, call the existing saveDraft() function."

Deliverables

  • Completed the mode selection drill (Task 1) — all 6 scenarios answered correctly
  • Compared two models on the same prompt (Task 2) — noted differences in style and quality
  • Practiced iterative refinement with 4 follow-up messages (Task 3)
  • Wrote a structured prompt using Context → Task → Constraints (Task 4)
  • Used the "Explain Then Do" pattern for a form implementation (Task 5)
  • Identified and rewrote 3 anti-pattern prompts (Task 6)

Success Criteria

  • You can confidently choose between Ask, Edit, and Agent mode for any given task
  • You know when to switch models and why
  • Your prompts follow the Context → Task → Constraints structure
  • You use iterative refinement instead of mega-prompts
← Back to lessonTake the Quiz →