docs(guide): state the html tier's tag spelling on react-pages, scope the react tier's

[claude]

Fixes #7006

content/docs/guide/react-pages.md is the home for both source-authoring tiers, but the only tag-naming rule it stated was the react tier's PascalCase one. A kind:'html' author who followed the stated rule was rejected at save time, with nothing on the page naming the rule they broke — while the page's own html examples already demonstrated the correct spelling.

Prose only. No parser, manifest or authorable-surface change: the rule already is what this page now says, the page was simply silent about it. Modelled on the same fix upstream (objectstack-ai/objectstack#13734), which repairs this defect on content/docs/ui/react-pages.mdx.

(Tag names below are written without their angle brackets — the platform's body sanitizer eats short angle-bracket fragments.)

The rule, re-measured (not taken from the card)

Re-read on the objectstack checkout at a4816a7:

• packages/sdui-parser/src/index.ts:41 — const allowedTags = new Set(Object.keys(manifest.components));
• packages/sdui-parser/src/parse.ts:95 — if (this.opts.allowedTags && !this.opts.allowedTags.has(tag)) {, error is not an allowed component

An exact Set.has against the manifest's own keys. grep -nE 'toLowerCase|toUpperCase|localeCompare' over packages/sdui-parser/src/ hits only codegen.ts (which derives the react tier's PascalCase names) — nothing on the whitelist path. The card's premise holds.

Registry figures re-measured independently in this repo, from packages/core/src/registry/public-blocks.ts: 57 entries, 28 carrying a namespace prefix, 4 snake_case after the colon (record:related_list, record:line_items, record:quick_actions, record:reference_rail) — matching the card's numbers exactly.

Edit 1 — the kind:'html' section now states the spelling

Unchanged sentence above it: "Only tags in the public block manifest are allowed, props are validated against each block's declared inputs, and unknown tags are a hard error at save time."

Added after it:

Those tags are the registered type names, written verbatim — whatever the registry spells, character for character, including a record: / page: / element: / action: namespace prefix and any underscore inside the name: list-view, object-form, record:related_list, record:quick_actions. The whitelist is an exact string comparison, so nothing is normalised for you: the PascalCase tags this page shows for kind:'react' are the other tier's convention and are not registered names (ListView is rejected with ListView is not an allowed component), and neither is a name re-spelled to look uniform — record:related-list is not registered, only record:related_list is.

The wording deliberately says "the registered type name, verbatim" and never "kebab-case". "kebab-case" would send an author to record:related-list, which is not registered either — so the paragraph names that wrong spelling explicitly and rules it out. The upstream model PR does use the phrase "lowercase and kebab-case"; that half of its wording is not copied here.

Edit 2 — the react-tier naming passages are scoped

Before (scope table row):

| The public data blocks | Every public non-container block, as a PascalCase tag — but what resolves and what you author against are two different sets, below. |

After:

| The public data blocks | Every public non-container block, as a PascalCase tag on this tier — but what resolves and what you author against are two different sets, below. |

Before (tag derivation):

Tags are derived by splitting the registry type on -, _ and : and PascalCasing each part: object-grid → ObjectGrid, record:details → RecordDetails.

After:

On this tier tags are derived by splitting the registry type on -, _ and : and PascalCasing each part: object-grid → ObjectGrid, record:details → RecordDetails. A kind:'html' page writes the registry type itself instead — object-grid, record:details.

The section names kind:'html' at seven places; only the section itself gained the rule.

Reverse verification — anchor counts (grep -c, on the page)

Phrase	Count	Expected
registered type names, written verbatim	1	1 (new anchor)
On this tier	1	1 (new anchor)
on this tier (case-insensitive, both edits)	2	2
record:related-list (the wrong spelling, named as wrong)	1	1
kebab-case — control, the trap wording	0	0
PascalCase	2	was 1; +1 in the html section, where it is named as the other tier's convention

Gates (redirect-then-capture; each gate's own verdict line)

Run at 7686d3c80, worktree base b8e3f8379.

Gate	Exit	Verdict line
node scripts/check-doc-fence-languages.mjs	0	check:doc-fences — every TypeScript block in 227 document(s) is fenced ts/tsx/typescript…
node scripts/check-doc-component-types.mjs	0	Every documented component type is registered.
node scripts/check-doc-snippet-types.mjs	0	Semantic phase: 467 of 467 block(s) judged, 0 failed. / Every covered documentation snippet compiles against the built types.
node scripts/check-doc-links.mjs	0	Links are valid across 17 scan roots.
node scripts/check-control-bytes.mjs	0	check-control-bytes: OK (scanned 6407 tracked text file(s); skipped 85 binary).
node scripts/check-changeset-presence.mjs	0	No source or published contract of a released package changed in this range, so no changeset is owed.
node scripts/check-governed-queue-guard.mjs --test content/docs/guide/react-pages.md	0	NOT GOVERNED — 1 path(s) checked against 5 governed surface(s); none matched.

check:doc-snippets was run for real, not narrowed: its --build-filter closure (34 tasks) was built first under the shared verify lock, so its exit 0 is a verdict about documents rather than an exit 2 "could not run".

Extra control-byte sweep beyond the gate: grep -naP '[\x00-\x08\x0b\x0c\x0e-\x1f\x7f]' over the edited file — no hits.

Files changed

content/docs/guide/react-pages.md — 1 file, +18 / −5. No changeset (the gate above says none is owed: a docs page under content/docs releases nothing).

---

Generated by Claude Code