What problem does it solve? LLM responses arrive as free-form text that breaks when fed directly into code, causing malformed JSON, missing fields, and wrong types in production pipelines. This Skill provides schema-constrained extraction patterns across OpenAI, Anthropic, and Google APIs so LLM output can safely feed database writes, API calls, and UI rendering. ## Core Features & Use Cases - Provider-Specific Methods: Covers OpenAI's response_format with JSON Schema and strict mode, Anthropic's tool_use blocks for structured extraction, and Gemini's responseSchema with responseMimeType. - Schema-First Workflow: Guides defining Pydantic models or Zod schemas with field-level descriptions, then validating responses with model_validate() or .parse() before downstream use. - Retry and Validation Logic: Includes patterns for retry loops on validation failures, handling refusals, enum casing mismatches, and empty array edge cases. - Use Case: You are building an invoice processing pipeline where Claude must return vendor names, totals, and line items as typed objects. This Skill shows how to force tool_use output, extract the tool_use block, and validate it before writing to your database. ## Quick Start Ask the AI to extract structured data from text using OpenAI structured outputs or Anthropic tool_use with a Pydantic schema matching your required fields.
