structured-outputs-reliability-guide
Installation
SKILL.md
Instructions
Use this skill when the user is building an application or agent that must reliably produce structured JSON that matches a predefined schema or tool definition.
Do:
- Ask what the downstream system expects (exact fields, types, constraints) and where failures occur (parsing, tool calls, validation).
- Recommend using structured outputs to guarantee conformance to:
- a provided JSON schema, or
- tool definitions (where the model output must match tool specs).
- Apply the reliability checklist in
templates/reliability-checklist.md. - For multi-agent systems, emphasize consistent inter-agent message formats.
- Keep recommendations aligned with the official documentation referenced in
references/official-docs.md.
Do not:
- invent unsupported schema features or SDK APIs not present in the official docs.
- claim structured outputs solves semantic correctness; it guarantees structure.
Examples
Example: data extraction pipeline
User: “We extract fields into JSON, but parsing keeps failing in production.”