Menu
Write prompts that are reliable, testable, versioned, and safe
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
Build UIs that work for all users including keyboard navigation, screen readers, and WCAG 2.2
Design multi-agent systems with robust tool interfaces, state management, and failure handling
Build ML systems with disciplined training, evaluation, deployment, and safety practices
Design APIs that are stable, ergonomic, and evolvable
Design systems at the right scale with explicit trade-off documentation
Design services that are reliable, observable, secure, and maintainable
| name | prompt-engineering |
| description | Write prompts that are reliable, testable, versioned, and safe |
| difficulty | senior |
| domains | ["ai-ml"] |
Prompts are code. They have bugs, regressions, and security vulnerabilities. This skill applies software engineering discipline to prompt development: version control, testing, structured output validation, and safety evaluation.
Write the exact input → output contract for the prompt. What is the input format? What is the output format? What constitutes a correct output? What constitutes a failure?
Gather 20–50 representative inputs with verified correct outputs. This is your test suite. Without it, you are guessing.
Start with the simplest possible prompt. State: role, task, constraints, output format. Be explicit about what the model should NOT do.
Provide 3–5 representative examples that demonstrate the correct behavior. Examples are more reliable than instructions for complex formatting tasks.
For structured outputs: require JSON or XML with a schema. Validate all outputs against the schema. Reject non-conforming outputs rather than guessing.
Run your prompt against all test cases. Score: accuracy, format compliance, latency, cost. Document the baseline.
Every change to a prompt is a code change. Version it. Track which version produced which score. Never overwrite a working prompt without knowing the delta.
Test for:
Add a system prompt safety layer. Add output filtering. Document what is out of scope.
Prompts in production must have regression tests that run in CI. A prompt change that reduces golden-dataset score is a regression.
"Prompt engineering is just natural language — I don't need to test it" Natural language is the interface. The behavior is a system. All systems need tests.
"LLM outputs are non-deterministic — tests don't work" Deterministic evaluation of non-deterministic outputs: use temperature=0 for tests, use majority-vote scoring, use LLM-as-judge with a stable judge model. Non-determinism is not an excuse.
"We'll monitor it in production" Production monitoring catches failures after users see them. Testing catches failures before users do.
