prompt-engineering

Topics in the source

• Prompt Engineering
• Quick Start
• When to Use
• Best Practices
• Framing
• Structured Output
• Reasoning
• Dependencies

Designing prompts that reliably steer LLM behavior. Getting consistent structured output from LLMs Reducing hallucinations in critical workflows Making prompts robust across models and versions Building prompts for agents and tools Give the model a clear role and task State constraints explicitly (format, length, tone) Provide context before asking the question Ask for one thing per prompt when possible Ask for JSON/YAML with a schema Use few-shot examples of the desired format Request self-checking: "validate your JSON before responding" Prefer responseformat/tools when the API supports them Use chain-of-thought for multi-step problems Ask for step-by-step reasoning before the final answer Keep reasoning separate from the final deliverable Avoid leaking internal reasoning into user-facing output 1. Define the task, expected output, and failure modes. 2. Write a role + constraints prompt; include an output schema. 3. Add 2-3 few-shot examples of the exact format. 4. For reasoning tasks, request step-by-step with separated final answer. 5. Test against a held-out set; measure accuracy and format compliance. 6. Iterate: add failing cases, tighten constraints, trim noise. 7. Version p

This is an excerpt. Refer to the complete source and its current requirements.