Prompt Engineering & Structured Output

C3 · 20% of the exam · 6 lessons · about 56 min of reading

Prompts that hold up in production: system prompts, XML tags, few-shot examples, step-by-step reasoning, JSON schemas, tool-forced output, validation and retry.

Course progress0/6

Short by the site author (n_tech_am) on YouTube, in Armenian. Pressing play loads it from youtube-nocookie.com. ·open on YouTube

System prompts and rolesUse the system parameter for the role and standing rules, keep task data in the user turn, and replace old prefill tricks with clear instructions.9 min
XML tags for instructions, context and dataWrap each part of a prompt in its own descriptive tag, put long data first and the question last, and ask for answers inside tags you can parse.9 min
Few-shot examplesShow Claude three to five relevant, varied examples wrapped in tags to pin down format and judgement, and avoid examples that teach the wrong pattern.8 min
Step-by-step reasoning and thinkingLet Claude reason before it answers with adaptive thinking and effort, know when manual chain-of-thought still applies, and when to split work into a chain of calls.10 min
JSON schemas and tool-forced outputGet machine-readable JSON with structured outputs, strict tools and tool_choice, and know which option each model accepts.10 min
Validation and retry loopsCheck every structured answer against stop_reason, schema and business rules, feed specific errors back for a bounded retry, and escalate when retries run out.10 min