CCA-F: Claude Certified Architect — Foundations

By Anthropic · Question Mode

QuestionQ96

Prompt Engineering & Structured Output

You are developing a structured-data extraction system with Claude. The system pulls information from unstructured documents, validates its output with JavaScript Object Notation (JSON) schemas, and preserves high accuracy. It must manage edge cases gracefully and integrate with downstream systems.

Your schema contains a skills: string[] field. Production monitoring identifies three consistency problems:

  1. Compound phrases such as "Python and SQL" are sometimes retained as a single entry and sometimes split.
  2. Implied but unmentioned skills occasionally appear in extracted results.
  3. Similar documents yield dramatically different array lengths (5–10 versus 40+ entries).

Your prompt currently states, "Extract all skills mentioned."

What is the most effective improvement?

Explanation

Few-shot examples establish the intended extraction policy by showing how compound skills are represented, requiring an explicit textual mention, and illustrating the appropriate level of entry granularity. Relevant, diverse examples—including edge cases—improve Claude’s extraction accuracy and consistency more reliably than a vague instruction alone. An arbitrary maximum count does not define the correct extraction boundary and can exclude valid skills.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!