Reference Skills

Reference skills carry background knowledge an agent draws on while working with LAI or with a Langium project. They are marked user-invocable: false, so they're not intended to be run directly. Instead, an agent should load them as context, and use those skills to inform how it reasons about your project throughout a session. You can think of these as instructions and reference material that helps the actionable skills and the CLI workflow skills work more effectively (without duplicating the same information in each skill).

Both reference skills ship in the skills/ directory and are installed alongside everything else via npx skills add eclipse-langium/langium-ai (see Install).

lai

A skill that functions as a guide to using the lai CLI: commands, configuration, the evaluation workflow, and result analysis. It mirrors the CLI loop with per-step detail on lai.config.jsonc, the language descriptor, the system prompt, and the .eval.ts files. It also includes details about how a provider is wired up in evals/utils.ts.

When an agent should use it: any time it's running lai commands, editing a descriptor, writing or debugging evaluation files, or trying to understand what a given step of the loop produces and why. It's the map to consult before the actionable lai-gen-* skills will make sense.

Source: skills/lai/SKILL.md

langium

A comprehensive reference for how Langium projects work, covering grammar definition, code generation, the dependency-injection system, lexing and parsing, the document lifecycle, scope computation and name resolution, linking and cross-references, validation, workspace management, and LSP integration. It's the largest skill in the set and is useful well beyond LAI. It's a general Langium primer an agent can rely on for any Langium DSL work.

When an agent should use it: whenever it needs to understand or modify the language itself. This includes: reading a grammar, tracing how a document is parsed and validated, debugging scoping or linking, or explaining a language feature accurately. The generation skills lean on this knowledge to describe a DSL correctly in a descriptor, prompt, or eval.

Source: skills/langium/SKILL.md

Where they fit

The two reference skills underpin the actionable skills and thread through the whole typical workflow. An agent keeps them in context while it generates and refines the descriptor, system prompt, evals, and MCP server.