guidance

Constrain LLM output with regex and grammars for structured JSON, XML, and code.

2|2|Updated Apr 16, 2026
One-click install
npx skills add https://github.com/huidge/hermes-skills --skill guidance-huidge
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: guidance
Source: https://github.com/huidge/hermes-skills/tree/main/mlops/inference/guidance
Command: npx skills add https://github.com/huidge/hermes-skills --skill guidance-huidge

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Regex and grammars to control LLM output, guaranteeing structured, valid JSON/XML/code results and reducing invalid generations.

Core Features & Use Cases

  • Regex and CFG-based constrained generation for reliable outputs.
  • Token healing and grammar-driven validation to prevent formatting errors.
  • Multi-step workflows and orchestration for complex prompts across local or cloud backends.

Quick Start

Generate a constrained JSON object for a user profile using Guidance with a JSON grammar.

Frequently Asked Questions about guidance

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I constrain LLM output to ensure valid JSON?

You can constrain LLM output to ensure valid JSON by applying regex and context-free grammars. This enforces structured generation rules, guaranteeing valid JSON or XML results and preventing formatting errors.

What is constrained generation and how does it work with regex?

Constrained generation uses regex and grammar-based validation to control LLM outputs. It restricts the generation process to valid patterns, ensuring structured results and reducing invalid generations across code and data formats.

Can I use grammars to constrain generation across local and API backends?

Yes, you can apply grammar-based constrained generation across local and API backends. This allows you to orchestrate multi-step workflows and guarded prompts consistently, regardless of your chosen backend environment.

Does constrained generation work for multi-step workflows and code generation?

Yes, constrained generation supports multi-step workflows and code generation by applying grammars to control LLM output. This ensures structured, valid results and orchestrates complex prompts across various backends.

Do I need the Guidance library to prevent LLM formatting errors with token healing?

Yes, you need the Guidance library to prevent formatting errors with token healing. It provides grammar-driven validation and guarded prompts to ensure precise, structured outputs and reduce invalid generations.