Prompt Engineering

Writing effective prompts is the most important skill for getting good results from AI coding tools.

Core principles

Be specific

Bad:

Fix the bug

Good:

Fix the TypeError in src/api/users.ts:42 where `user.email` is undefined
when the OAuth provider doesn't return an email field

Provide context

Include relevant details the AI needs:

What you're trying to accomplish
Where in the codebase the change should happen
Why the current behavior is wrong
How you expect it to work after the fix

Break down complex tasks

Instead of one large request, split into steps:

First, analyze the existing authentication flow
Then, propose changes to support SSO
Finally, implement the changes

Prompt templates

Bug fix

In [file], the [function/component] has a bug where [describe the problem].
Expected behavior: [what should happen].
Actual behavior: [what happens instead].
Fix this while maintaining [any constraints].

New feature

Add [feature] to [component/module].
It should [describe behavior].
Follow the existing patterns in [reference file/module].

Code review

Review [file or diff] for:
- Security issues
- Performance concerns
- Error handling gaps
- Adherence to our coding standards

Refactoring

Refactor [function/class] to [goal].
Keep the public API the same.
Make sure all existing tests still pass.

Anti-patterns to avoid

Vague requests — "Make this better" gives vague results
Too many goals at once — Focus on one thing at a time
No acceptance criteria — Always define what "done" looks like
Ignoring context — The AI can't read your mind; share what you know