🤖 AI Engineering · Prompt Engineering

Guarantee structured outputs from LLMs

Get machine-parseable JSON reliably using schema-first prompting, constrained decoding, and repair loops.

intermediate~35 minAI EngineersML EngineersLLM App Developers

Steps

1Provide the JSON Schema in the prompt and ask for schema-valid output only
2Prefer provider-native structured-output or function-calling modes when available
3Set temperature low for extraction; reserve high temperature for ideation tasks
4Validate with a strict parser; reject unknown keys and wrong types
5On failure, re-ask once including the validation error verbatim
6Quarantine repeated failures to a dead-letter store for rubric fixes

Common Pitfalls

▲Parsing with regex instead of a real schema validator
▲Accepting markdown-fenced JSON silently breaking downstream parsers

Commands

Install with skills CLI

$ npx skills add aniruddhaadak80/skills --skill prompt-engineering-structured-output

Install globally

$ npx skills add aniruddhaadak80/skills --skill prompt-engineering-structured-output -g

Tags

#json#structured-output#reliability#ai-engineering#prompt-engineering

Install this skill

$ npx skills add aniruddhaadak80/skills --skill prompt-engineering-structured-output -a opencode

SKILL.md

# Guarantee structured outputs from LLMs

> Get machine-parseable JSON reliably using schema-first prompting, constrained decoding, and repair loops.

## When to Use This Skill

Get machine-parseable JSON reliably using schema-first prompting, constrained decoding, and repair loops.

## Steps

1. Provide the JSON Schema in the prompt and ask for schema-valid output only
2. Prefer provider-native structured-output or function-calling modes when available
3. Set temperature low for extraction; reserve high temperature for ideation tasks
4. Validate with a strict parser; reject unknown keys and wrong types
5. On failure, re-ask once including the validation error verbatim
6. Quarantine repeated failures to a dead-letter store for rubric fixes

## Common Pitfalls

- Parsing with regex instead of a real schema validator
- Accepting markdown-fenced JSON silently breaking downstream parsers

View source on GitHub ↗

Organize identity, rules, tools, and output contracts into sections that evolve without breaking behavior.

🤖 AI Engineering·~30m

Open →

Select few-shot examples that move accuracy (Classification tasks)intermediate

Choose and order examples covering edge cases so the model generalizes instead of copying surface patterns.

🤖 AI Engineering·~30m

Open →

Select few-shot examples that move accuracy (Extraction tasks)intermediate

Choose and order examples covering edge cases so the model generalizes instead of copying surface patterns.

🤖 AI Engineering·~30m

Open →