Skip to main content
← Back to list
01Issue
FeatureClosedSwamp ClubPublic
AssigneesNone

Relationships

#3185 Docs: state the literal type/version catalog contract and the uncatalogued-model push warning in the extensions publishing reference

Opened by stack72 · 10/8/2026

Follows swamp-club#2486.

content/manual/reference/extensions/publishing.md should:

  • State that the registry catalog reads a model's type and version as string literals from a plain export const model = { ... } object. Models built by a shared factory and spread into the export must repeat both literals there, and the export must not carry a type annotation (use satisfies after the object instead). A model that does not meet this still installs and runs, but the registry does not list it, so it is missing from extension search --content-type models and extension info contentMetadata.models.
  • List the new warning-level rule uncatalogued-model next to bare-specifiers in the push warnings section: it reports each model entry point the catalog would leave out and what could not be read (a non-plain export, a missing literal type, or a missing literal version). Like bare-specifiers it cannot be accepted. It appears in extension push (including --dry-run and --json) and extension quality, and changes no exit code.
02Bog Flow
✓OPEN○TRIAGED○IN PROGRESS◉CLOSED

Closed

10/8/2026, 6:34:03 PM

No activity in this phase yet.

03Sludge Pulse
Editable. Press Enter to edit.

stack72 commented 10/8/2026, 6:34:03 PM

Documented in swamp-club PR #1314 (merged). content/manual/reference/extensions/publishing.md has a new Model catalog subsection: the registry catalog reads type and version as string literals from a plain export const model = { ... } object, a factory-built model repeats both literals in the export, and the export carries no type annotation (use satisfies). A model that does not meet this installs and runs but is missing from extension search --content-type models and contentMetadata.models. The uncatalogued-model rule is listed next to bare-specifiers in the review warnings and as not acceptable in the acceptance rules table. The warning text comes from this issue; the rule was not in a released CLI when the docs were written, so no sample output is shown yet.

Sign in to post a ripple.