@canonical/pragma-cli 0.32.0 → 0.34.0

package/pragma.conf.ts

@@ -0,0 +1,664 @@
+/**
+ * The pragma distribution config — identity, default packs, and the read
+ * stories those packs supply.
+ *
+ * Consumed three times: statically imported by `src/kernel/config/defaults.ts`
+ * as the lowest config layer (compiled into the binary — no fs), by
+ * `src/capabilities/distribution.ts` (which compiles the declared stories into
+ * capability modules at module load), and at build time by the bundler. A
+ * NON-magic name on purpose: `findProjectConfig` only discovers
+ * `pragma.config.{ts,js}`, so this file is never mistaken for a project config.
+ * Validated by the same `parseRawConfig` as every layer.
+ *
+ * INERT DATA. It must never gain a value import: `src/constants.ts` statically
+ * imports it to project the program's identity, so anything that runs at this
+ * module's load runs on `--help`, `__complete` and `--version`. That also makes
+ * the reverse edge impossible — importing `constants.js` from here is a
+ * temporal-dead-zone cycle. It is also why no string below names the binary
+ * except `name` itself: an `emptyRecovery.cli` is the command WITHOUT the
+ * binary name, and the renderer prepends the consuming distribution's.
+ * `capabilities/lazy.test.ts` pins both halves of that.
+ */
+
+import type { RawConfig } from "./src/kernel/config/types.js";
+import type { PackDefinition } from "./src/kernel/packs/types.js";
+
+/**
+ * The design-system domain colophon — the ontology + graph story, surfaced by
+ * `pragma colophon` after pragma's own. Declared on the flagship UI-block noun
+ * because `block` most embodies the design system; it narrates the DOMAIN (what
+ * the graph models), not the toolchain (pragma's built-in section).
+ */
+const DESIGN_SYSTEM_COLOPHON = `The Canonical design system is a **knowledge graph**, not a component library.
+Every block, token, modifier, standard, and tier is a node in an RDF store,
+described by the \`ds:\` ontology and queried the same way whether you reach it
+over GraphQL or raw SPARQL.
+
+## What the graph models
+
+- **Blocks** — components, patterns, layouts, and subcomponents (the \`ds:UIBlock\`
+  family). A block carries its anatomy, guidelines, and \`when to use\` / \`when
+  not to use\` narrative as graph properties, not prose in a wiki.
+- **Modifiers** — families of variant values (\`ds:hasModifierFamily\` →
+  \`ds:hasModifier\`) a block composes, so a variant is a relationship, not a
+  string.
+- **Tokens** — the themeable design values, resolved per theme.
+- **Standards** — the do / don't coding guidance, categorized and linked to the
+  blocks they govern.
+
+## How it fits together
+
+- **Tiers** are a hierarchy (\`global\` > \`apps\` > \`apps/lxd\`): a lower tier
+  inherits and overrides the blocks of its ancestors, so scoping a query to a
+  tier walks that chain.
+- **Channels** (\`normal\`, \`experimental\`, \`prerelease\`) gate visibility, so an
+  in-progress block never leaks into a stable answer.
+
+## Why RDF
+
+One graph makes every relationship first-class and queryable: \`block lookup\`
+follows edges to modifiers and subcomponents, \`graph query\` runs arbitrary
+SPARQL, and \`ontology show\` reads the schema itself. The store is built once
+by \`sources update\` and addressed by content hash, so the domain you query is
+exactly the domain that was published.`;
+
+/**
+ * The read stories the design-system pack supplies — `block`, `token`,
+ * `modifier` and `tier` as declared data rather than code.
+ *
+ * The association with the pack is the point: when `@canonical/design-system`
+ * starts shipping its own `stories/*.json`, this block is deleted and the
+ * package tier takes over with no code change here or anywhere else. The local
+ * `readonly PackDefinition[]` annotation is what type-checks them —
+ * `PackSource.stories` is deliberately `readonly unknown[]`, because the config
+ * layer does not know the pack grammar (`parsePackDefinition` does).
+ */
+const designSystemStories: readonly PackDefinition[] = [
+  // `block list` is declared content (L-OPEN-9): one unfiltered SELECT over
+  // the four UIBlock classes, listing ALL blocks — experimental and alpha ones
+  // included, for everyone — until filtering returns in declared form. The
+  // hand-written tier-chain/channel filtering (and its `--all-tiers` escape)
+  // is removed with the code, an owner-signed consequence. Display parity
+  // lives IN the query: `VALUES ?class` closes the type set, so the BINDs
+  // that derive `name` (declared name, else the IRI's local name), `type`
+  // (lowercased class local name), and `tier` (tier IRI's local name) operate
+  // over known shapes — the `standard` story's BIND/COALESCE precedent.
+  //
+  // `block lookup` is served over the GraphQL fetch path (ONE generated document
+  // over the `UIBlock` interface covering Component/Pattern/Layout/Subcomponent).
+  // Base level mirrors the old summary view (name/tier/summary); the default is
+  // `detailed`, matching the old CLI which rendered anatomy and modifiers
+  // without a flag. A derived name that maps onto no schema field is omitted
+  // (OPTIONAL parity), so a graph lacking whenToUse/whenNotToUse degrades
+  // gracefully.
+  //
+  // Disclosure declares the FULL canonical ladder `[summary, standard,
+  // detailed]` — the same set `standard` declares — so a config
+  // `detail=standard` names a level `block` advertises rather than one it
+  // silently accepts-then-degrades. Block carries no `standard`-tier content of
+  // its own; gating is by canonical index, so `standard` resolves to the base
+  // view. Only the level SET is aligned: the per-noun DEFAULT stays domain-tuned
+  // (`block` rich-by-default, `standard` terse-by-default).
+  {
+    noun: "block",
+    description: "List all design system blocks.",
+    toolDescription:
+      "List all design system blocks with their type, tier, and modifier families. Use when browsing which blocks exist. Example: block_list {}.",
+    colophon: DESIGN_SYSTEM_COLOPHON,
+    list: {
+      query: [
+        "SELECT ?uri ?name ?type ?tier",
+        '       (GROUP_CONCAT(DISTINCT ?modName; separator=", ") AS ?modifiers)',
+        "WHERE {",
+        "  VALUES ?class { ds:Component ds:Pattern ds:Layout ds:Subcomponent }",
+        "  ?uri a ?class .",
+        "  OPTIONAL { ?uri ds:name ?dsName }",
+        "  OPTIONAL { ?uri ds:tier ?tierUri }",
+        "  OPTIONAL { ?uri ds:hasModifierFamily ?family . ?family ds:name ?modName }",
+        '  BIND(COALESCE(?dsName, REPLACE(STR(?uri), "^.*[/#]", "")) AS ?name)',
+        '  BIND(LCASE(REPLACE(STR(?class), "^.*[/#]", "")) AS ?type)',
+        '  BIND(REPLACE(STR(?tierUri), "^.*[/#]", "") AS ?tier)',
+        "}",
+        "GROUP BY ?uri ?name ?type ?tier",
+        "ORDER BY ?name",
+      ].join("\n"),
+      columns: [
+        { field: "name", label: "Name" },
+        { field: "type", label: "Type" },
+        { field: "tier", label: "Tier" },
+        { field: "modifiers", label: "Modifiers" },
+        { field: "uri", label: "IRI" },
+      ],
+      emptyRecovery: {
+        message:
+          "No blocks in the store. Build it from the configured design-system packs.",
+        cli: "sources update",
+      },
+    },
+    lookup: {
+      source: "graphql",
+      toolDescription:
+        'Get detailed information about one or more design system blocks including anatomy, modifiers, and properties. Use when you need the full spec of specific blocks by name — detail: "summary" trims to the base view. Example: block_lookup { names: ["Button"] }.',
+      by: "ds:name",
+      types: ["ds:Component", "ds:Pattern", "ds:Layout", "ds:Subcomponent"],
+      graphqlType: "UIBlock",
+      fields: [
+        { name: "tier", property: "ds:tier", label: "Tier" },
+        {
+          name: "figmaLink",
+          property: "ds:figmaLink",
+          label: "Figma",
+          level: "detailed",
+        },
+      ],
+      sections: [
+        { name: "summary", property: "ds:summary", label: "Summary" },
+        {
+          name: "whenToUse",
+          property: "ds:whenToUse",
+          label: "When to use",
+          level: "detailed",
+        },
+        {
+          name: "whenNotToUse",
+          property: "ds:whenNotToUse",
+          label: "When not to use",
+          level: "detailed",
+        },
+        {
+          name: "guidelines",
+          property: "ds:guidelines",
+          label: "Guidelines",
+          level: "detailed",
+        },
+        {
+          name: "anatomyDsl",
+          property: "ds:anatomyDsl",
+          label: "Anatomy (DSL)",
+          kind: "code",
+          level: "detailed",
+        },
+        {
+          name: "anatomyClassic",
+          property: "ds:anatomyClassic",
+          label: "Anatomy (classic)",
+          kind: "code",
+          level: "detailed",
+        },
+      ],
+      expand: [
+        {
+          name: "modifierFamilies",
+          heading: "Modifier Families",
+          relation: "ds:hasModifierFamily",
+          level: "detailed",
+          select: [
+            { name: "name", property: "ds:name" },
+            {
+              name: "values",
+              relation: "ds:hasModifier",
+              select: [{ name: "name", property: "ds:name" }],
+            },
+          ],
+        },
+        {
+          name: "properties",
+          heading: "Properties",
+          relation: "ds:hasProperty",
+          kind: "table",
+          level: "detailed",
+          select: [
+            { name: "name", property: "ds:name" },
+            { name: "type", property: "ds:propertyType" },
+            { name: "optional", property: "ds:optional" },
+          ],
+        },
+        {
+          name: "subcomponents",
+          heading: "Subcomponents",
+          relation: "ds:hasSubcomponent",
+          level: "detailed",
+          select: [
+            { name: "name", property: "ds:name" },
+            { name: "uri", property: "ds:name", graphqlField: "uri" },
+          ],
+        },
+      ],
+      disclosure: {
+        levels: ["summary", "standard", "detailed"],
+        default: "detailed",
+      },
+      sample: {
+        fixedCount: true,
+        toolDescription:
+          "Return randomly selected complete design-system blocks as exemplars. Use BEFORE writing queries to see actual data shapes, anatomy, and property names.",
+      },
+    },
+  },
+
+  // Design tokens: SPARQL-sourced on both verbs. There is no `ds:Token` GraphQL
+  // type to project against when the graph ships no tokens, and the lookup reads
+  // a property path (`ds:tokenType/rdfs:label`) only SPARQL can express. The
+  // `emptyRecovery` install hint is the story users see on an empty store. The
+  // noun is now purely declarative: `token add-config` wrote a starter file, and
+  // L-OPEN-9 removed it rather than growing the read grammar a mutation verb.
+  {
+    noun: "token",
+    description: "List all design tokens.",
+    toolDescription:
+      "List all design tokens with their type. Use when browsing which tokens exist under the active scope. Example: token_list {}.",
+    list: {
+      query: [
+        "SELECT ?uri ?name ?category WHERE {",
+        "  ?uri a ds:Token ;",
+        "       ds:tokenId ?name .",
+        "  OPTIONAL {",
+        "    ?uri ds:tokenType ?type .",
+        "    ?type rdfs:label ?category .",
+        "  }",
+        "}",
+        "ORDER BY ?name",
+      ].join("\n"),
+      columns: [
+        { field: "uri", label: "IRI" },
+        { field: "name", label: "Name" },
+        { field: "category", label: "Type" },
+      ],
+      emptyRecovery: {
+        message:
+          "No tokens in the store. Build it from the configured design-system packs.",
+        cli: "sources update",
+      },
+    },
+    lookup: {
+      source: "sparql",
+      by: "ds:tokenId",
+      type: "ds:Token",
+      toolDescription:
+        'Get type and theme values for one or more design tokens by name. Use when resolving specific tokens\' light/dark values. Example: token_lookup { names: ["color.primary"] }.',
+      fields: [
+        {
+          name: "category",
+          property: "ds:tokenType/rdfs:label",
+          label: "Type",
+        },
+        { name: "valueLight", property: "ds:valueLight", label: "Light value" },
+        { name: "valueDark", property: "ds:valueDark", label: "Dark value" },
+      ],
+      sample: {
+        fixedCount: true,
+        toolDescription:
+          "Return randomly selected complete design tokens (with theme values) as exemplars. Use BEFORE writing queries to see actual data shapes.",
+      },
+    },
+  },
+
+  // Modifier families. `modifier list` is SPARQL (an alternation path collects
+  // values asserted in either direction); `modifier lookup` is GRAPHQL, where
+  // the compiled `ModifierFamily.modifiers` field is the declared-inverse union
+  // and resolves both directions in ONE generated document — exactly the case
+  // the GraphQL fetch path exists for.
+  {
+    noun: "modifier",
+    description: "List all modifier families.",
+    toolDescription:
+      "List all modifier families with their values. Use when browsing which modifier families exist and the values each allows. Example: modifier_list {}.",
+    list: {
+      query: [
+        "SELECT ?uri ?name",
+        '       (GROUP_CONCAT(DISTINCT ?valueName; separator=", ") AS ?values)',
+        "WHERE {",
+        "  ?uri a ds:ModifierFamily ;",
+        "       ds:name ?name .",
+        "  OPTIONAL {",
+        "    ?uri (ds:hasModifier|^ds:modifierFamily) ?value .",
+        "    ?value ds:name ?valueName .",
+        "  }",
+        "}",
+        "GROUP BY ?uri ?name",
+        "ORDER BY ?name",
+      ].join("\n"),
+      columns: [
+        { field: "uri", label: "IRI" },
+        { field: "name", label: "Name" },
+        { field: "values", label: "Values" },
+      ],
+      emptyRecovery: {
+        message:
+          "No modifier families in the store. Build it from the configured design-system packs.",
+        cli: "sources update",
+      },
+    },
+    lookup: {
+      source: "graphql",
+      by: "ds:name",
+      type: "ds:ModifierFamily",
+      toolDescription:
+        'Get values and usage details for one or more modifier families by name. Use when you need the allowed values of specific families. Example: modifier_lookup { names: ["importance"] }.',
+      expand: [
+        {
+          name: "values",
+          heading: "Values",
+          relation: "ds:hasModifier",
+          select: [{ name: "name", property: "ds:name" }],
+        },
+      ],
+      sample: {
+        fixedCount: true,
+        toolDescription:
+          "Return randomly selected complete modifier families (with value lists) as exemplars. Use BEFORE writing queries to see actual data shapes.",
+      },
+    },
+  },
+
+  // Tiers. The hierarchy is encoded in the slash-separated path string
+  // (`apps/lxd`), not in graph edges, so `tier list` is a flat, name-ordered
+  // list. `tier lookup` resolves a tier by its declared name and expands the
+  // blocks scoped DIRECTLY to it through the inverse path `^ds:tier` — a
+  // single-hop expand the generated sub-SELECT emits as `<tier> ^ds:tier ?child`
+  // (property paths, inverse included, are in-contract for `PackExpand.relation`).
+  // Retiring the bespoke single-`<name>` lookup for this declared variadic one
+  // is an L-OPEN-9 covenant change, recorded in the covenant's $comment.
+  {
+    noun: "tier",
+    description: "List all tiers in the design system ontology.",
+    toolDescription:
+      "List all tiers in the design-system ontology. Use when understanding the tier hierarchy before setting a tier filter. Example: tier_list {}.",
+    list: {
+      query: [
+        "SELECT ?uri ?name WHERE {",
+        "  ?uri a ds:Tier ;",
+        "       ds:name ?name .",
+        "} ORDER BY ?name",
+      ].join("\n"),
+      columns: [
+        { field: "uri", label: "IRI" },
+        { field: "name", label: "Name" },
+      ],
+    },
+    lookup: {
+      source: "sparql",
+      by: "ds:name",
+      type: "ds:Tier",
+      description: "Show tiers by name, with the blocks scoped to each.",
+      toolDescription:
+        'Get one or more tiers by name, with the blocks scoped directly to each. Use when you need which blocks a specific tier carries. Example: tier_lookup { names: ["apps/lxd"] }.',
+      expand: [
+        {
+          name: "blocks",
+          heading: "Blocks",
+          relation: "^ds:tier",
+          select: [{ name: "name", property: "ds:name" }],
+        },
+      ],
+    },
+  },
+];
+
+/**
+ * The read story the code-standards pack supplies — `standard` as declared data.
+ *
+ * Normalized for the v2 grammar: the old `digest` level is the canonical
+ * `standard`, so disclosure gates by the canonical index. The default is
+ * `summary` (base fields by name), `--detail standard` adds the `dos` examples,
+ * `--detail detailed` adds `donts`. `cs:extends` stays the raw IRI in JSON
+ * (renderers compact it at display time).
+ */
+const codeStandardsStories: readonly PackDefinition[] = [
+  {
+    noun: "standard",
+    description: "List all code standards.",
+    toolDescription:
+      "List code standards. Optionally filter by category or search term.",
+    list: {
+      query: [
+        "SELECT ?uri ?name ?category ?description",
+        "WHERE {",
+        "  ?uri a cs:CodeStandard ;",
+        "       cs:description ?description .",
+        "  OPTIONAL { ?uri cs:name ?n . }",
+        '  BIND(COALESCE(?n, REPLACE(STRAFTER(STR(?uri), "#"), "\\\\.", "/")) AS ?name)',
+        "  OPTIONAL {",
+        "    ?uri cs:hasCategory ?cat .",
+        "    ?cat cs:slug ?category .",
+        "  }",
+        "}",
+        "ORDER BY ?name",
+      ].join("\n"),
+      columns: [
+        { field: "uri", label: "IRI" },
+        { field: "name", label: "Name" },
+        { field: "category", label: "Category" },
+        { field: "description", label: "Description" },
+      ],
+      filters: [
+        {
+          param: "category",
+          variable: "category",
+          description: "Filter by category name.",
+        },
+      ],
+      search: {
+        variables: ["name", "description"],
+        description: "Search in name and description.",
+      },
+    },
+    verbs: [
+      {
+        verb: "categories",
+        description: "List all standard categories with counts.",
+        toolDescription: "List all code standard categories.",
+        query: [
+          "SELECT ?name (COUNT(?standard) AS ?count)",
+          "WHERE {",
+          "  ?cat a cs:Category ;",
+          "       cs:slug ?name .",
+          "  OPTIONAL {",
+          "    ?standard a cs:CodeStandard ;",
+          "              cs:hasCategory ?cat .",
+          "  }",
+          "}",
+          "GROUP BY ?name",
+          "ORDER BY ?name",
+        ].join("\n"),
+        columns: [
+          { field: "name", label: "Category" },
+          { field: "count", label: "Standards" },
+        ],
+      },
+    ],
+    lookup: {
+      source: "sparql",
+      by: "cs:name",
+      type: "cs:CodeStandard",
+      description:
+        "Look up detailed information for a standard by name, IRI, or glob.",
+      toolDescription:
+        "Get detailed information about one or more code standards including dos and donts with code examples. Address standards by name, prefixed name (cs:…), absolute IRI, or glob pattern (react/component/*).",
+      fields: [
+        {
+          name: "category",
+          property: "cs:hasCategory/cs:slug",
+          label: "Category",
+        },
+        {
+          name: "description",
+          property: "cs:description",
+          label: "Description",
+        },
+        { name: "extends", property: "cs:extends", label: "Extends" },
+      ],
+      expand: [
+        {
+          name: "dos",
+          heading: "Do",
+          relation: "cs:do",
+          select: [
+            { name: "caption", property: "cs:description" },
+            { name: "language", property: "cs:language" },
+            { name: "code", property: "cs:code" },
+          ],
+          level: "standard",
+        },
+        {
+          name: "donts",
+          heading: "Don't",
+          relation: "cs:dont",
+          select: [
+            { name: "caption", property: "cs:description" },
+            { name: "language", property: "cs:language" },
+            { name: "code", property: "cs:code" },
+          ],
+          level: "detailed",
+        },
+      ],
+      disclosure: {
+        levels: ["summary", "standard", "detailed"],
+        default: "summary",
+      },
+      sample: {
+        description:
+          "Return randomly selected complete standard instances as exemplars for shape discovery.",
+        toolDescription:
+          "Return 1–5 randomly selected complete code standard instances as exemplars. Use BEFORE writing queries to see actual data shapes, property names, and value formats. Each call returns different instances.",
+      },
+    },
+  },
+];
+
+export default {
+  name: "pragma",
+  help: "Explore the design system",
+  // The toolchain's own colophon — CONTENT this distribution declares, not
+  // machinery: `pragma colophon` renders whatever stands here as its first
+  // section, titled with the distribution's name, before any active pack's
+  // domain colophon. A fork tells its own story by editing this declaration.
+  // `markdown` is the full narrative; `summary` is the condensed `--format llm`
+  // form. Both are BODIES with no leading H1 (the renderer supplies the
+  // heading), grounded in this tree's real architecture.
+  colophon: {
+    markdown: `pragma is a **domain-based toolchain**: one CLI and one MCP server projected
+from a single grammar, serving a knowledge-graph domain that reads as data.
+
+## The effect monad
+
+Reads are plain \`async\` functions; a mutation instead *describes* its effects
+as a \`Task\` (\`@canonical/task\`) that is interpreted — under the real node
+interpreter, a \`--dry-run\` planner, or an \`--undo\` reverser. Describe-then-
+interpret means dry-run and undo come for free, and the dispatcher tells the two
+worlds apart on one bit: \`capability.mutates\`.
+
+## One grammar, many projections
+
+Every capability is a \`VerbSpec\` — a noun, its params, its effect profile, its
+formatters. The CLI commands, the MCP tools, shell completion, and the
+surface/docs are all *projections* of that one shape, so they cannot drift. The
+projected surface is frozen in a covenant (\`surface/surface.v2.json\`): a single
+source of truth a test asserts the live grammar still emits, tool for tool.
+
+## LLM-optimized output
+
+Each verb renders three ways — \`plain\` for a terminal, \`json\` for the machine
+envelope, and \`llm\` for condensed Markdown. \`--format llm\` (or a non-interactive
+stdout) selects the agent form: the same data, shaped for a model to read. This
+colophon is itself a showcase of that render model.
+
+## Modular, storeless by construction
+
+Capabilities ship as **modules** — named bundles of verbs with optional
+boot / resource / prompt hooks. A verb declares whether it \`needsStore\`, and
+the dispatcher boots the triple store *only* for those; a storeless verb
+(\`info\`, \`config\`, \`capabilities\`, \`colophon\`) never pays for the graph.
+
+## Scaffolding
+
+\`pragma create\` scaffolds components, packages, and applications through the
+\`@canonical/summon-*\` generators, reusing summon's rich Ink wizard when it runs
+interactively.
+
+## The domain reads as data
+
+A domain is a **pack**: a declarative \`PackDefinition\` (its list / lookup
+queries) compiled into verbs, backed by a content-addressed graphpack that
+\`sources update\` builds once. Swap the pack and the same pragma serves a
+different domain — including the domain colophon printed below this one.
+
+Made by the Canonical Webteam — https://canonical.com.`,
+    summary: `pragma is a domain-based toolchain: one CLI + MCP server projected from a single \`VerbSpec\` grammar.
+
+- **Effect monad** (\`@canonical/task\`): reads are async; a mutation returns an interpreted \`Task\`, so \`--dry-run\` and \`--undo\` are free. The dispatcher branches on \`capability.mutates\`.
+- **One grammar, many projections**: CLI, MCP tools, completion, and docs all project one \`VerbSpec\`; the emitted surface is frozen in a covenant so the projections never drift.
+- **LLM-optimized output**: every verb renders \`plain\` / \`json\` / \`llm\`; \`--format llm\` (or a piped stdout) emits condensed Markdown for agents.
+- **Modular + storeless**: capability modules; the triple store boots only for \`needsStore\` verbs.
+- **Domain as data**: a pack is a declarative \`PackDefinition\` compiled to verbs over a content-addressed graph built by \`sources update\`.
+
+Made by the Canonical Webteam — https://canonical.com.`,
+  },
+  issuesUrl: "https://github.com/canonical/pragma/issues",
+  packs: [
+    {
+      name: "@canonical/design-system",
+      source: "git+https://github.com/canonical/design-system.git#main",
+      stories: designSystemStories,
+    },
+    {
+      name: "@canonical/anatomy-dsl",
+      source: "git+https://github.com/canonical/anatomy-dsl.git#main",
+    },
+    {
+      name: "@canonical/code-standards",
+      source: "git+https://github.com/canonical/web-code-standards.git#main",
+      stories: codeStandardsStories,
+    },
+  ],
+  // This distribution's domain namespaces, declared once and read twice.
+  //
+  // (1) They pin what `sources update` builds the store with. The design system
+  // declares `ds:` twice — `…/` in `definitions/` and `…/data/` in `data/` —
+  // and prefix harvesting is last-wins over a filename sort, so which one binds
+  // is an accident of file naming. The config layer wins every harvest: without
+  // this pin an added or renamed upstream file can silently compact every `ds:`
+  // entity to the wrong prefix, and `block list` stops resolving.
+  //
+  // (2) They are the domain half of the kernel's `DEFAULT_PREFIX_MAP`
+  // (`src/kernel/render/prefixes.ts`) — what the CLI compacts in output AND
+  // what it expands a user-typed `ds:Button` / `cs:rule` through before a
+  // lookup. Drop a namespace here and its lookups stop resolving; that is why
+  // `cs:` is listed even though nothing rebinds it upstream.
+  prefixes: {
+    ds: "https://ds.canonical.com/",
+    cs: "http://pragma.canonical.com/codestandards#",
+  },
+  channel: "normal",
+  detail: "standard",
+} satisfies RawConfig;
+
+/**
+ * The domain terms the kernel reads this distribution's graph with.
+ *
+ * A separate export because it is NOT a config layer field: it is compiled in
+ * and read at module load by `src/kernel/vocabulary.ts`, whose readers (the
+ * storeless completion fast path, the pack index builder) cannot reach a config
+ * layer at all. Layering it would let a project config set it and change
+ * nothing. A fork edits these values and rebuilds; `parseVocabulary` type-checks
+ * the shape and rejects a term that is not a prefixed name.
+ *
+ * Every prefix used here must be bound in `prefixes` above. `rdfs:label` and
+ * `rdfs:comment` are deliberately absent — the kernel treats standard
+ * vocabulary as universal.
+ *
+ * The `prompt` terms are a READ CONTRACT, not a claim about instances: this
+ * distribution's graph currently carries no `ds:Prompt` entities at all, so
+ * `prompt list` is honestly empty. Declaring the shape anyway is what makes an
+ * empty result mean "the graph has none" rather than "nothing was declared".
+ */
+export const vocabulary = {
+  altName: "ds:name",
+  prompt: {
+    type: "ds:Prompt",
+    body: "ds:promptBody",
+    argument: "ds:promptArgument",
+    argName: "ds:argName",
+    argRequired: "ds:argRequired",
+  },
+};