Slate v2 Declaration Merging Recovery

Goal

Recover the legacy declaration-merging typing contract in package code first, then remove the example-side bandaids that only existed because the package surface drifted.

Source Of Truth

• .agents/rules/repair-drift.mdc
• legacy slate CustomTypes fixtures under /Users/zbeyens/git/slate/packages/slate/test/interfaces/CustomTypes
• current package surfaces under:
  • /Users/zbeyens/git/slate-v2/packages/slate
  • /Users/zbeyens/git/slate-v2/packages/slate-react
  • /Users/zbeyens/git/slate-v2/packages/slate-history

Phases

1. Audit legacy declaration-merging contract vs current package exports/types
2. Recover package-side aliases and merge seams in slate
3. Recover package-side composition helpers in slate-react / slate-history
4. Remove downstream bandaids in examples/tests
5. Verify package, site, and targeted browser proof

Findings

• Legacy CustomTypes covered more than Editor / Element / Text; it also covered Node, Point, Range, Selection, and Operation.
• Current docs/ledgers explicitly classify CustomTypes as a hard cut, so this work will need doc cleanup after the code truth is restored.
• Current RenderElementProps is nongeneric, so narrowing element.type does not narrow the whole props object in examples like check-lists.

Progress

• Phase 1 complete: legacy fixtures and current package seams loaded
• Phase 2 complete: slate now restores base aliases and the ExtendedType seam for editor/element/text/point/range/selection/operation contracts
• Phase 3 complete: slate-react restores package-side custom-types augmentation and shared render-element typing; slate-history now composes from BaseEditor again
• Phase 4 complete: removed a first batch of example-side casts and prop-shape bandaids that only existed because the package type surface drifted
• Phase 5 current read:
  • package build/typecheck is green for slate, slate-history, slate-react
  • site typecheck is green
  • focused browser batch is green for:
    • embeds
    • iframe
    • images
    • inlines
    • mentions
    • search-highlighting
  • paste-html still has one red Playwright row, but the failure is a strict locator ambiguity in the test (textbox.locator('code') resolves to three nodes), not a typing regression from this recovery
  • check-lists still has separate checkbox behavior debt; that proof lane was not closed by this typing recovery