Exam Q&A Generator

Generates a JSON question-and-answer bank from Microsoft Learn training material (Markdown files produced by the microsoft-exam-docs skill. Generally you will look for a CONTENT.md file and a SUMMARY.md file).

Artifact policy

Create exactly one file: the final question-and-answer JSON file named {exam-code}.json in the training directory.

When you are generating the file, make sure to remove any helper scripts that you generated. Where possible and applicable, use Powershell, Bash, jq, fd, rg, node, bun, and python. Preference goes for Node since you can write the validation rules in as part of the generation and verify the output.

Efficiency guidelines

• Build a quick content outline from headings before generating questions. Use it to identify learning paths, modules, units, and sections with enough body content.
• Allocate question counts per section before generation starts. This avoids over-generating and then discarding large batches.
• Generate in section-sized batches instead of repeatedly prompting over the entire CONTENT.md.
• When using subagents, pass only the relevant source section(s), the allocated question count, and the schema rules. Do not pass the entire exam content to every subagent unless the file is small.
• Prefer exact JSON arrays from generation batches so the coordinator can merge them directly with minimal rewriting.
• Do one final global pass for duplicates, topic consistency, answer/option integrity, IDs, and questionCount.
• Use inline validation commands where useful, but keep all intermediate state in memory and write only the final JSON file.