@tanstack/table-core

Keep private declarations out of the emitted .d.ts files, so consumers emitting declarations no longer fail with TS2883 (e.g. on RangeValue)

Improve tree-shaking further. The core no longer bundles the column ordering, column pinning, or row expanding code it used for leaf column order, pinned header order, and row display order, so a minimal table is about 0.27 kB (brotli) smaller. Features no longer bundle code from other features a table doesn't register: column sizing, column ordering, and cell selection without column pinning or column visibility, column pinning without column visibility, row pinning without row expanding, the paginated row model without row expanding, and the faceted unique and min/max values without global filtering. Without globalFilteringFeature, global facets now include no columns, the same as the filtered row model. Memoized row, column, header, and cell APIs also allocate less: each memo no longer creates an onAfterUpdate wrapper it doesn't need. table.getOrderColumnsFn(), which column ordering already assigned, is now declared on its table type.

The devtools Features panel sizes are regenerated for these changes.

Improve tree-shaking. Importing one built-in filterFn_*, sortFn_*, or aggregationFn_* now bundles only that function instead of its whole module. The core row model, header building, and the filtered, sorted, and grouped row models no longer import code from features a table doesn't register, so a minimal table is about 0.6 kB (brotli) smaller and grouping without rowAggregationFeature no longer bundles the aggregation executor. Add table.autoResetSorting(), and fix the Table_RowPagination type to declare the autoResetPageIndex method the table actually has (it was typed as _autoResetPageIndex).

The devtools Features panel now shows measured minified + brotli sizes for every feature, row model, and built-in function, and its total accounts for code that features share.

Refactor bundled Intent skills into smaller entry points with references loaded for the current task. Keep core feature architecture and shared state directly discoverable, and move individual features, adapter compositions, and detailed migration guidance into references.

Consumers with individual skill permissions or explicit agent mappings must replace retired feature/composition IDs with their owning entry points and refresh Intent mappings. See the Agent Skills guide for the replacement paths.

Update TanStack Store dependencies to the latest compatible patch releases.

Skip missing core rows when rebuilding a worker row-model payload so stale worker results cannot crash the table