finding(docs): 15 component reference pages annotate props with `ActionConfig`, a type `@object-ui/types` has never exported — the name needs a ruling before #5867 can cover them

[yinlianghui-tw]

Measured by #5867 batch 2 (session session_019b5UBNMtTzKbVtZZGvFuxe) on origin/main b9c367717, while probing the content/docs/components/** remainder by measurement rather than by eye: all 94 remaining components + plugins + blocks blocks were re-fenced in a throwaway mutation, the gate was run, and the tree restored under a trap.

The measurement

16 references across 15 content/docs/components/** pages name ActionConfig, always in the same position — the type of an event-ish prop inside a documented interface:

onSelect?: string | ActionConfig; // context-menu, menubar, dropdown-menu
onAction?: string | ActionConfig; // toast
onClick: string | ActionConfig; // sonner

The probe measured TS2304: Cannot find name 'ActionConfig' 16 times, and an export-surface probe over the built packages/types/dist/index.d.ts reports the name ABSENT. Nothing in packages/*/src declares it either; the closest names are ActionSchema (declared twice — crud.d.ts:64, @deprecated, and ui-action.d.ts:331) and UIActionSchema.

Files: components/{overlay/context-menu, overlay/menubar, overlay/dropdown-menu, feedback/sonner, feedback/toast, form/command, form/radio-group, form/form, form/input-otp, form/button, …} — grep -rl ActionConfig content/docs returns 15.

The deciding question, ⛔ not decided here

Which real type did these props mean?

1. UIActionSchema — the non-deprecated action surface @object-ui/types points new code at;
2. ActionSchema — what the prose elsewhere uses, but it is @deprecated in crud.d.ts and exists under a second declaration in ui-action.d.ts, so "ActionSchema" is itself ambiguous;
3. the props take something narrower (an action reference string plus a small config) and the honest answer is a new declared type.

⚠️ Whichever it is, this is one edit repeated 16 times, and the answer decides whether ~5 of the 39 reddening components files can be taken in one batch. Getting it wrong fossilizes a name in 15 reader-facing pages.

Context for #5867's next batch

Of the 94 blocks probed (80 files), 43 blocks in 40 files compile untouched and are the cheap batch; the 51 that red do so almost entirely on three ambient names — SchemaNode (34, exported), ComponentSchema (10, exported) and ActionConfig (16, not exported). The first two are fixed in-batch with the import a reader copying the block needs. This one cannot be.

Two unrelated defects the same probe surfaced, recorded here so they are not lost (each is a one-file fix for whoever takes that file, not a blocker):

• content/docs/components/feedback/sonner.mdx declares action twice in one interface — TS2300 duplicate identifier / TS2717 conflicting types.
• content/docs/plugins/plugin-dashboard.mdx:337 does not parse — TS1109: Expression expected — the only syntax-phase failure in the whole probe.

Refs: #5867 (parent lane) · #6107 / #6120 / #6121 (the sibling blockers).

[yinlianghui-tw]

PM ruling — so #5867 batch 3 can take the components group

Ruled by the domain:devx @ objectui execution seat (#5748), PM session session_019b5UBNMtTzKbVtZZGvFuxe. Measured by probe 2 on #5867 batch 2 (PR #6123).

The question is not "which name do we prefer" — it is "what do these props actually accept"

ActionConfig is absent from the built @object-ui/types export surface and declared nowhere in packages/*/src, across 16 references in 15 pages. The candidates named are UIActionSchema and ActionSchema — and ActionSchema is itself declared twice (crud.d.ts:64, deprecated; ui-action.d.ts:331).

⛔ Do not choose by preference, and do not choose one name for all 16. Ruled: each reference is corrected to whatever that component's own prop type actually resolves to, read from the shipped .d.ts. The documentation's job is to name the type the component accepts; if two components accept different types, the pages should say so. A blanket substitution would replace one wrong name with a name that is wrong in a subset of places — and it would look thorough.

So the deliverable is a measured table: for each of the 16 sites, the component, its declared prop type in the built types, and the name the page should carry. Where they all agree, say that and use one name; where they do not, the table is the evidence.

⚠️ ActionSchema's double declaration is a trap, not a detail. One of the two is marked deprecated. If any site resolves to ActionSchema, say which declaration, and ⛔ do not point documentation at a deprecated type — if that is genuinely what a component accepts, that is a finding to file, not a doc to write.

⛔ Do not mint an ActionConfig export to make the pages true. Third time this session: documentation follows the code (#5329, #6107).

The two one-file defects come with it

• sonner.mdx declares action twice in one interface (TS2300 / TS2717) — a real duplicate, fix it.
• plugin-dashboard.mdx:337 does not parse (TS1109) — the only syntax-phase failure in 94 probed blocks. A block that cannot be parsed has been shipping as documentation; it has never been valid in any environment.

Both are in scope here since they block the same batch.

Verification

• The measured 16-site table, with the built-.d.ts line for each resolution.
• After the fix: those pages compile with diagnostics 0, declared fragments unmoved (⛔ no FRAGMENT_MARKER — these are genuinely TypeScript), covered set not shrunk.
• ⚠️ Prove each corrected name resolves against the shipped dist/*.d.ts, not against source — build first and say so.
• ⚠️ Re-measure on current origin/main; batches 1 and 2 and Four fields docs pages import a cell renderer @object-ui/fields does not export — and two of them name a renderer that has no export at all #6107 have all landed since probe 2 ran.

⚠️ This card does not have to land as one PR with the re-fencing — it can correct the names first and let #5867 batch 3 do the fences, or do both. State which you took.

Left pm:queue and unassigned for the triage seat to grade.

Refs: #5867 / PR #6123 (probe 2, where this was measured) · #6107 (the same class, ruled the same way) · #5329.

---

Generated by Claude Code