prompt-engineer

Provide your goal and expected output format, and the Skill runs a short adaptive interview before building the prompt. For N8N it loads agent-specific patterns, keeps the prompt concise, and enforces structured output such as strict JSON for downstream nodes.

Use the MELHORAR mode: paste your existing prompt and the Skill diagnoses it across clarity, structure, techniques, anti-patterns, and platform fit. It then returns an annotated before/after rewrite with a score comparison showing what changed and why.

The Skill applies official Anthropic techniques: clarity and specificity, XML tag structuring, Chain of Thought, Few-Shot examples, Role Prompting, prefilling, prompt chaining, long-context optimization, and hallucination prevention. The references file details all 13 techniques with examples.

Not directly. The Skill includes a multi-model adaptation table: GPT-4 lacks native XML support and prefilling, Gemini has partial XML support, and Llama needs more explicit Chain of Thought. It adapts techniques per target model when asked.

Hallucinations usually come from missing grounding instructions. The Skill adds anti-hallucination patterns such as permission to say "I don't know", citation requirements, and explicit flagging of assumptions, which sharply reduce invented answers.

Use chaining when a task has three or more distinct steps, conditional branches, or when one step's output feeds the next. The Skill detects when complexity exceeds a single prompt and recommends splitting into a chained sequence with XML handoffs.