Constrain LLM generation to produce valid JSON, XML, or code.
npx skills add https://github.com/devMoez/titan --skill guidance-devmoez
Please help me install this Agent Skill. Skill: guidance Source: https://github.com/devMoez/titan/tree/main/optional-skills/mlops/guidance Command: npx skills add https://github.com/devMoez/titan --skill guidance-devmoez
Guidance solves the problem of unreliable LLM outputs by enforcing constrained generation so results always match a required syntax such as valid JSON, XML, or code-like formats.
Use Case: Generate an API payload where every field is syntactically valid (e.g., timestamps, numeric ranges, enumerated categories), then pass it directly into downstream systems without validation failures.
Ask an AI to generate a person object as valid JSON using Guidance while constraining the name with a regex and the email with an email-pattern grammar.
High-intent search queries and answers about installing and using this skill.
Constrained generation enforces specific syntax rules like valid JSON during decoding, ensuring LLM outputs always match required formats without retry loops. This eliminates parsing failures by guaranteeing structural validity at generation time.
Regex-constrained generation forces LLM outputs to match specific patterns like emails, dates, or IDs. This ensures individual fields in structured outputs adhere to exact formatting rules during decoding rather than relying on post-generation validation.
Multi-step workflow orchestration combines stateful generation functions with constrained decoding to execute multi-stage pipelines while keeping outputs well-formed. This approach maintains format validity across complex agent loops and tool interactions.
Constrained decoding with selectable enumerations restricts LLM generation to predefined categories, ensuring classification outputs always match fixed options. This removes validation failures by enforcing valid selections during generation itself.
Grammar-constrained generation supports full context-free grammars to enforce nested structure rules during LLM decoding. This guarantees syntactically valid complex outputs like nested JSON or XML without requiring post-hoc parsing checks.
kaminocorpConstrain LLM outputs with regex and grammar-based generation.
microwindValidate and format JSON data with syntax, structural, and optional schema checks.
matthew-johnsonApply Guidance constraints to generate validated JSON, XML, or code.
ruiyangruiyiGenerate schema-constrained LLM outputs as JSON, regex matches, or Pydantic instances.
