Constrain LLM generation to valid JSON, XML, and code via regex and grammar rules.
npx skills add https://github.com/qcmuu/AI-Research-Skills --skill guidance-qcmuu
Please help me install this Agent Skill. Skill: guidance Source: https://github.com/qcmuu/AI-Research-Skills/tree/main/16-prompt-engineering/guidance Command: npx skills add https://github.com/qcmuu/AI-Research-Skills --skill guidance-qcmuu
Guidance helps you control an LLM’s output so it always matches required formats, preventing invalid or malformed responses when you need machine-readable results like JSON, XML, or code.
Use cases include: extracting well-formed fields (emails, dates, IDs), generating valid API payloads, building structured data pipelines, and implementing constrained agent steps that depend on valid intermediate artifacts.
Ask the AI to generate a valid JSON object by constraining each field with a regex so the output is guaranteed to parse.
High-intent search queries and answers about installing and using this skill.
To force valid JSON without retries, apply grammar-based decoding to constrain token generation. This enforces syntax rules during inference, guaranteeing machine-readable structures and eliminating fragile post-processing.
Token healing fixes structured generation errors by improving boundary consistency between prompt text and generated tokens. This technique prevents formatting glitches and spacing errors, ensuring constrained outputs maintain strict syntactic validity.
To extract structured data fields using regex validation, restrict LLM generation to specific patterns. This ensures fields like emails, dates, and IDs are syntactically valid upon generation, suitable for structured data extraction pipelines.
Yes, the Guidance framework supports multi-step agent workflows by composing constrained steps using Pythonic functions and stateful logic. This enables agentic pipelines to depend on valid intermediate artifacts without format errors.
Yes, you can use constrained generation with existing transformers models via backend-compatible execution. The Guidance framework integrates with transformers to enforce regex patterns and grammar rules directly during the model's token generation process.
You should not use grammar-based decoding when unconstrained natural language generation is required, as it restricts outputs to strict patterns. It is specifically designed for structured data extraction and production pipelines where format errors are costly.
ovachieverConstrain LLM outputs to valid JSON, XML, and code using regex and grammar constraints.
peteromalletConstrain LLM outputs with regex and grammars using Guidance.
shuff57Constrain LLM outputs with regex and grammars for structured JSON, XML, and code generation.
carterwayneskhizeineGenerate schema-constrained JSON, XML, regex, or Pydantic outputs from LLMs.
