glossary · git:20261027.bbe68e5 · 2026-10-27 · sha256 a6593a781100dd42
glossary git:20261027.bbe68e5A
Immutable. This exact content never changes and is served at /api/v1/blob/a6593a781100dd42.
No tool to install first
claude plugin marketplace add https://markdownregistry.com/mp/a/art_5sxennfiu55zmuea/a6593a781100dd42ba65b77ad75913feb32934e0d6875adbd527f85072b8ddc9.json
Then /plugin install glossary@markdownregistry-5sxennfiu55zmuea in Claude Code. It installs this exact version, the one graded A here, pinned to the commit it was graded at, and leaves it there.
--- name: glossary description: >- Writes the ubiquitous language of a bounded context, before any code, and keeps its CONTEXT.md, the per-context document every agent reads first. One row per term with its meaning here and what it must not be confused with. Use when a business term is missing, ambiguous or used with two meanings, when the generator or language/glossary-term refuses a name, when a CONTEXT.md is empty or stale, or as the glossary step of a new project. For the boundary itself use bounded-contexts. argument-hint: "[context]" allowed-tools: - Read - Write - Edit - Bash(npx ddd-ts-kit list *) - Bash(npx ddd-ts-kit docs) - Bash(npx ddd-ts-kit check *) --- # Ubiquitous language Context: `$ARGUMENTS` (every context of `docs/architecture/contexts.md` when empty) The glossary is the source of every class, method, event, command and permission name. The generator refuses a name whose words it does not define, and `language/glossary-term` fails the check. A refusal class is named from it too (`SlotTaken`: `Slot`, `Take`), with the qualifiers of a refusal besides: `invalid`, `unknown`, `already`, `too`, `no`, `missing`, `required` (`BookingTooLate`, `InvalidAppointment`). ## Where it is written | Project state | File | |---|---| | before the bootstrap | `docs/architecture/glossary.md`, one section per context; the bootstrap turns each row into a `--term "<Term>=<meaning>\|<not to be confused with>"`, three cells kept, and deletes the file | | after | `src/contexts/<context>/CONTEXT.md`, section `## Ubiquitous language`; there is no second copy | ## A context a preset writes A context that `generate context <name> --preset <preset>` will write (today `identity`) takes the preset's glossary as its draft, not an empty one: its terms are in [identity-preset.md](identity-preset.md), next to this skill, the rows the preset writes into its CONTEXT.md word for word (before the bootstrap the project has no CLI to read them from). Keep the preset's word for each concept it names (`Organisation` is the tenant, `Member` a membership), and add only the business terms beyond it, one row each. The preset's terms win: the glossaries downstream refer to them. ## Collect the terms From the discovery timeline (or from the owner), for one context at a time: - each **noun** the business uses: candidates for aggregates, entities, value objects; - each **verb**: candidates for commands and intention methods (`place`, `cancel`, `issue`); - each **fact**, in the past tense: candidates for events (`placed`, `cancelled`); - each **role** that acts. For each, one sentence of meaning **in this context**, and for a word a neighbour also uses, what it must not be confused with there. Refuse technical words (`record`, `data`, `item`, `manager`, `info`, `status` alone) and ask for the business word. One word per concept: pick `cancel` or `void`, not both. Ten precise terms beat fifty vague ones. ```markdown ## ordering | Term | Meaning | Not to be confused with | |---|---|---| | Order | A customer's request to buy products, from placement to delivery | billing: Invoice, the debt it creates | | Place | To submit an order for fulfilment; it cannot be edited afterwards | | ``` ## After the bootstrap: `CONTEXT.md` It is what an agent reads instead of the code of the context. Keep it true and short. | Section | Written by | Content | |---|---|---| | `## Purpose` | you | two sentences in business terms; the first one feeds the context map | | `## Classification` | `generate context --classification <c> --reason "..."` | the classification and one sentence of why | | `## Ubiquitous language` | you | the table above | | `## Aggregates`, `## Published events`, `## Permissions`, `## Scheduled jobs`, `## Background jobs`, `## State machines`, `## Process managers`, `## Relationships`, `## Reference` | `npx ddd-ts-kit docs`, between markers | never edit between the markers; add below a list one line per aggregate with the invariant it guards | A new term enters with the code that needs it: `npx ddd-ts-kit generate <kind> ... --term "<Term>=<meaning>"` adds the row in the same command; `--term "<Term>=<meaning>|<not to be confused with>"` fills the third cell too, for homonyms only (a cell never holds a `|`, so it cannot clash with a meaning). Finish with `npx ddd-ts-kit check --rule tree/context-doc --rule language/glossary-term`. ## Done when Put this to the owner: **would a domain expert sign each definition, and does every word of the timeline have exactly one meaning per context?** On a yes, the next step is `ddd-ts-kit:context-map`.