ubiquitous-language
Mine the live conversation for domain-relevant nouns, verbs, and concepts; resolve synonyms and overloaded terms into a canonical, opinionated glossary. Persist the result to UBIQUITOUS_LANGUAGE.md so subsequent sessions inherit the same vocabulary. Re-invocation refines the file in place rather than overwriting.
When a candidate term collides with usage already present in the codebase, dispatch an Explore agent (fd-first discovery, git grep/ast-grep content search) to confirm the dominant naming before recommending a winner. The user supplies domain intent; the codebase supplies factual usage.
Modality vs adjacent skills: This skill extracts glossary from raw conversation when no documented domain language exists yet. Domain-model grilling grills a plan against an already-documented CONTEXT.md/ADRs. Pick ubiquitous-language when you are creating the artifact; pick the grilling workflow when you are stress-testing one. The two compose: build the glossary here, then promote stable terms into CONTEXT.md and let the grilling workflow defend it thereafter.
Process
- Scan the dialogue for domain-relevant nouns, verbs, and concepts. Skip generic programming nouns (array, function, endpoint) unless they carry domain weight.
- Identify three failure modes: same word for different concepts (ambiguity), different words for the same concept (synonyms), vague or overloaded terms.
- Propose canonical terms with explicit aliases-to-avoid. Be opinionated — pick one winner per concept and justify briefly.
- Persist
UBIQUITOUS_LANGUAGE.mdto the working directory using the format inreferences/UBIQUITOUS-LANGUAGE-FORMAT.md. - Emit a short inline summary of additions, renames, and flagged ambiguities so the user can react in-thread.