| name | ubiquitous-language |
| description | Extract a domain glossary from the current dialogue; flag ambiguities, propose canonical terms, persist to `UBIQUITOUS_LANGUAGE.md`. Trigger when the user is hardening domain terminology, building a glossary, or fresh domain concepts surface in conversation without documented 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.md to the working directory using the format in references/UBIQUITOUS-LANGUAGE-FORMAT.md.
- Emit a short inline summary of additions, renames, and flagged ambiguities so the user can react in-thread.
Rules
- One sentence per definition. Define what the term is, not what it does.
- Group terms into multiple tables when natural clusters exist (subdomain, lifecycle, actor). Force no grouping when a single cohesive table reads cleanly.
- Express relationships with bold term names and explicit cardinality where obvious (
exactly one, one or more, optional).
- Flagged ambiguities require a recommendation, not a question. The skill resolves; the user overrides.
- Treat module and class identifiers as glossary candidates only when they carry domain meaning. Prefer abstract domain terms over implementation aliases.
- Include an example dialogue (3–5 exchanges) demonstrating the terms used precisely. The dialogue clarifies boundaries between adjacent concepts.
Re-invocation
When invoked again in the same conversation:
- Read the existing
UBIQUITOUS_LANGUAGE.md.
- Incorporate new terms surfaced since the last pass.
- Update definitions when shared understanding has shifted; preserve the old wording in commit history rather than the file body.
- Re-flag any new ambiguities; demote resolved ambiguities to a closed note or remove them once the glossary stabilises.
- Refresh the example dialogue so it exercises the latest term set.
Reference materials
references/UBIQUITOUS-LANGUAGE-FORMAT.md — table layout, relationship syntax, dialogue template, flagged-ambiguities format. Language-agnostic.