lai-gen-evals

Generate comprehensive .eval.ts evaluation suites for Langium DSL projects.

28|4|Updated Apr 10, 2025
One-click install
npx skills add https://github.com/eclipse-langium/langium-ai --skill lai-gen-evals
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: lai-gen-evals
Source: https://github.com/eclipse-langium/langium-ai/tree/main/skills/lai-gen-evals
Command: npx skills add https://github.com/eclipse-langium/langium-ai --skill lai-gen-evals

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill helps you create an evaluation suite that measures whether an LLM produces Langium DSL code that is syntactically correct, semantically valid, matches user intent, and handles edge cases—so you can iterate reliably instead of relying on superficial “it parses” checks.

Core Features & Use Cases

  • Expands eval coverage by category: organizes evals into focused suites like syntax, semantics, intent, edge cases, and understanding.
  • Leverages real Langium validation: uses LangiumEvaluator to validate generated programs against your actual parser and validator.
  • Supports robust scoring and debugging: guides you to compute meaningful 0–1 scores (including composites) and return useful diagnostics and extra metadata.

Quick Start

Use the lai-gen-evals skill to generate a comprehensive eval suite inside your project’s evals/ directory based on your existing descriptor, system prompt, and Langium language validation behavior.

Frequently Asked Questions about lai-gen-evals

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

FAQPage Schema
How do I build an LLM evaluation suite for a Langium DSL project?

You can generate an LLM evaluation suite by creating .eval.ts files in an initialized evals directory that use the langium-ai-tools eval API to test syntactic correctness, semantic validity, and user intent alignment.

What is semantic validation in LLM-generated DSL code?

Semantic validation verifies generated programs against your actual Langium parser and validator using LangiumEvaluator to ensure generated DSL code is logically valid and aligns with user intent.

How do I test intent matching for LLM outputs in a domain-specific language?

Test intent matching by expanding eval coverage into focused suites for understanding and edge cases, computing meaningful 0–1 composite scores, and returning useful diagnostics and extra metadata.

Do I need a configured generateResponse helper to run Langium evals?

Yes, running Langium evals requires a configured generateResponse helper, an initialized evals directory, and Langium services for evaluation to execute the LangiumEvaluator.

What's the best way to organize test suites for LLM model comparison in Langium?

The best way to organize test suites for model comparison is to categorize evals into focused suites like syntax, semantics, intent, edge cases, and understanding to build a reliable eval matrix.

Why does my LLM DSL evaluation only check if the code parses?

Your evaluation only checks parsing because it lacks expanded eval coverage for semantic validity and intent alignment, which requires generating refined eval files using the langium-ai-tools eval API.