dx-domain-discover

name	dx-domain-discover
description	Mine a brownfield codebase for its domain vocabulary and seed the project glossary.
disable-model-invocation	true
argument-hint	["module or path"]

dx-domain-discover

Bootstrap foundation/glossary.md from an existing codebase. Mine the code for the ubiquitous language already living in it — the terms the domain uses — and write them down so every later skill names things consistently. This is the one-time (per-module) extraction pass; /dx-domain keeps the glossary sharp during ongoing work. Re-runnable: pass [module or path] to scope the sweep to one area and grow the glossary incrementally.

Guard. foundation/glossary.md must exist (seeded by /dx-init) — if it is missing, tell the user to run /dx-init first, then stop. Read the current glossary before mining so you extend it, never clobber it.

Invoke dx-references with knowledge-layer for the glossary entry shape and the standards/lessons/glossary distinction — this skill writes the glossary and nothing else.

1 — Mine the vocabulary

Spawn built-in Explore subagents (fan-out, read-only), scoped to [module or path] when given, else the whole repo. Harvest candidate domain terms from:

• Identifiers — type/class/entity names, enums, key function and method names.
• Module and package names — the boundaries the code already draws.
• Comments and existing docs — READMEs, ADRs, doc-comments where terms get defined in prose.

Keep only terms specific to this domain. Drop general programming concepts (cache, retry, handler, DTO) — they are not ubiquitous language.

2 — Resolve and write

For each surviving term, write a glossary entry in the shape from knowledge-layer (**Term**: definition — domain-only, no implementation. _Avoid_: confusable term). Be opinionated: when several names map to one concept, pick one and list the rest under _Avoid_.

Surface, don't average. When you find a term clash (two definitions for one word) or a fuzzy/overloaded term, do not guess — list it and ask the user to resolve it, then record their decision. Keep the file glossary-ONLY: no implementation detail, no rationale, no decisions (those are lessons).

Done when

New and sharpened terms are in foundation/glossary.md and every clash was resolved by the user. Print a short summary — terms added, terms already present, clashes resolved — then:

Glossary seeded: foundation/glossary.md
Next: /dx-domain-discover <another-module>   — extend the sweep to another area
  or: /dx-domain                             — keep the glossary sharp as you work

Stop. Do not chain into another skill.