4. Prompt Engineering

4.1 What is a prompt?

A prompt is an instruction or input provided to an AI system.

Weak:

Fix this.

Better:

Investigate the failing login test.

Do not edit files yet.

Identify:
1. The failing test
2. The root cause
3. The relevant files
4. A proposed fix
5. How the fix should be verified

4.2 A useful prompt structure

Role / context
Goal
Scope
Constraints
Inputs
Expected output
Verification

Example:

Context:
This is a FastAPI application.

Goal:
Fix the failing authentication test.

Scope:
Only modify authentication-related code.

Constraints:
Do not change the database schema.

Verification:
Run the authentication test suite.

Output:
Explain the root cause and summarize the changes.

4.3 Planning prompts

For large tasks, ask the agent to analyze before modifying.

Analyze the repository and create an implementation plan.

Do not modify files.

Include:
- Current architecture
- Files that will change
- New files
- Risks
- Tests
- Rollback considerations

4.4 Implementation prompts

After approving a plan:

Implement the approved plan.

Follow AGENTS.md.

Keep the change scoped.

Run relevant tests.

Do not make unrelated refactors.

4.5 Review prompts

Review the current changes.

Do not modify files.

Check:
- Correctness
- Security
- Error handling
- Performance
- Tests
- Maintainability

Report findings with file and line references where possible.

4.6 Prompt anti-patterns

Avoid:

• Huge unrelated context
• Contradictory instructions
• Vague goals
• Missing verification
• Asking for irreversible actions without approval

Exercise

Write prompts for:

1. Code exploration
2. Architecture planning
3. Implementation
4. Testing
5. Security review