Slate v2 Udecode Open Issues Reproduction Ralplan

Verdict

Status: done.

There are 11 open issues in udecode/slate as of the live gh read. This plan accounts for all of them through exact fixes, already-accounted browser proof, or deliberately scoped non-claims. The issues cluster into four shared runtime owners:

• history and selection snapshots after replacement or undo: udecode/slate#9, #11, #12, #14
• clipboard and void serialization: udecode/slate#7, #10, #13
• composition cancellation and caret ownership: udecode/slate#6
• target ownership across voids, native controls, and iframes: udecode/slate#5, #8, #15

No Fixes claim is allowed until the original repro is red against local .tmp/slate-v2, green against the live reference or a written expected model, and then green after the implementation change. Rows that were already green stay already-accounted.

Current Evidence

Live issue intake used GitHub CLI:

gh issue list -R udecode/slate --state open --limit 1000 --json number,title,body,labels,url,updatedAt,createdAt,author

Current local source owners checked in .tmp/slate-v2:

• example server command exists as bun serve, serving site on port 3100 from .tmp/slate-v2/package.json
• history snapshots are owned by .tmp/slate-v2/packages/slate-history/src/with-history.ts
• beforeinput text replacement routes through .tmp/slate-v2/packages/slate-react/src/editable/editing-kernel.ts and .tmp/slate-v2/packages/slate-react/src/editable/mutation-controller.ts
• undo hotkeys route through .tmp/slate-v2/packages/slate-react/src/editable/keyboard-input-strategy.ts
• selection import/export is owned by .tmp/slate-v2/packages/slate-react/src/editable/selection-controller.ts
• composition events are owned by .tmp/slate-v2/packages/slate-react/src/editable/runtime-composition-events.ts and .tmp/slate-v2/packages/slate-react/src/editable/composition-state.ts
• target bridge is owned by .tmp/slate-v2/packages/slate-react/src/editable/runtime-target-bridge.ts
• clipboard export/import is owned by .tmp/slate-v2/packages/slate-dom/src/plugin/dom-clipboard-runtime.ts
• existing Playwright rows cover pieces of iframe, mentions, images, editable-voids, paste-html, richtext, and plaintext, but not the full issue repro set below

Intent And Boundary

Intent: turn every currently open udecode/slate issue into issue-first Slate v2 reproduction work, then execute the smallest architecture repair that fixes the shared runtime owner.

Outcome: every open issue has one of these states:

• fixes-claimed: exact original repro proved red, fixed, and verified
• improves-claimed: shared runtime behavior improves but exact repro is not closed
• related: pressure is accounted for but not in this pass
• needs-repro: issue body or video is insufficient for an implementation claim

In scope:

• udecode/slate#5 through #15, all currently open as of the gh read
• local reproduction against http://localhost:3100
• live reference comparison against https://www.slatejs.org
• .tmp/slate-v2 runtime, history, clipboard, selection, composition, and example tests once execution starts

Non-goals:

• no code patches before red proof
• no GitHub comments, labels, closes, or PR text changes
• no Fixes udecode/slate#... wording before exact proof exists
• no broad rewrite just because the bug list is ugly

Decision boundary:

• If a repro is green locally, mark it already-accounted with the command and browser evidence.
• If Playwright is green but Chrome/dev-browser reproduces the bug, trust the browser path and preserve the failed harness mismatch in the plan.
• If issue #13 cannot be reconstructed from its video, leave it needs-repro and do not infer a fix from the title.
• If a fix needs a runtime/API change, prefer the long-term raw Slate substrate over an example-local workaround.

Decision Brief

Principles:

• Issue-first, not vibes-first. The user gave open issues; each one gets a concrete repro.
• Runtime owners beat example patches. A bug that repeats across examples is a runtime contract failure until proven otherwise.
• Slate stays raw. Product/editor opinions belong in Plate unless the issue exposes a missing raw primitive.
• Browser behavior needs browser proof. IME, clipboard, iframe, void selection, and external paste cannot be closed by package-only tests.

Chosen approach:

1. Reproduce all 11 issues in four architecture clusters.
2. Add failing proof rows before implementation.
3. Fix the shared runtime owner for each cluster.
4. Sync issue claims only after the exact original repro is green.

Rejected:

• Fixing one example at a time. That would miss shared ownership bugs.
• Marking the plan done after issue intake. Reproduction is still missing.
• A full rewrite before red proof. That is just expensive guessing.
• Treating live reference divergence as automatically correct. The live behavior is the starting expectation, but the raw Slate contract still has to make sense.

Open Issue Matrix

Execution Plan

Phase 0: Harness Setup

Run from /Users/zbeyens/git/slate-v2:

bun serve

Use local URL http://localhost:3100. For Playwright rows, set:

PLAYWRIGHT_BASE_URL=http://localhost:3100

Do not treat a green synthetic row as browser closure for IME, external clipboard, or iframe toolbar behavior. Use dev-browser or real Chrome when the observable bug depends on browser state.

Phase 1: History Replacement And Undo

Issues: #9, #11, #12, #14.

Add red proof rows first:

• playwright/integration/examples/plaintext.test.ts
• playwright/integration/examples/richtext.test.ts
• playwright/integration/examples/hovering-toolbar.test.ts
• add or extend inlines, styling, code-highlighting, mentions, and tables rows as needed
• packages/slate-history/test/history-contract.ts for package-level selection-before, replacement, undo, and redo snapshots

Expected architecture repair:

• replacement over an expanded selection must be one history-owned transaction with the correct selectionBefore
• whole-document replacement must restore original children and selection on undo
• plain typing undo must restore caret at the edit point, not the visual line start
• marked replacement must not crash when beforeinput imports a DOM selection over formatted text

Likely files:

• packages/slate-history/src/with-history.ts
• packages/slate-react/src/editable/editing-kernel.ts
• packages/slate-react/src/editable/mutation-controller.ts
• packages/slate-react/src/editable/keyboard-input-strategy.ts
• packages/slate-react/src/editable/selection-controller.ts

Verification target:

bun test packages/slate-history/test/history-contract.ts
PLAYWRIGHT_BASE_URL=http://localhost:3100 playwright test playwright/integration/examples/plaintext.test.ts playwright/integration/examples/richtext.test.ts playwright/integration/examples/hovering-toolbar.test.ts --project=chromium

Phase 2: Clipboard And Void Serialization

Issues: #7, #10, #13.

Add red proof rows first:

• mentions selected inline void copy/paste and cut crash
• images selected block void copy to external/native clipboard payload
• paste-html exact video-derived input once reconstructed
• package proof in packages/slate-dom/test/clipboard-boundary.ts where the repro can be modeled without browser-only APIs

Expected architecture repair:

• inline void copy/cut/paste must not depend on a missing DOM attach target
• selected block void copy must place useful text/html and text/plain data on the clipboard for external apps
• intra-Slate fragment data must keep working while external HTML fallback is useful
• HTML paste import must fail closed or normalize instead of crashing

Likely files:

• packages/slate-dom/src/plugin/dom-clipboard-runtime.ts
• packages/slate-react/src/editable/runtime-clipboard-events.ts
• playwright/integration/examples/mentions.test.ts
• playwright/integration/examples/images.test.ts
• playwright/integration/examples/paste-html.test.ts

Verification target:

bun test packages/slate-dom/test/clipboard-boundary.ts
PLAYWRIGHT_BASE_URL=http://localhost:3100 playwright test playwright/integration/examples/mentions.test.ts playwright/integration/examples/images.test.ts playwright/integration/examples/paste-html.test.ts --project=chromium

Phase 3: IME Cancellation

Issue: #6.

Add red proof rows first:

• manual or dev-browser Chrome Chinese IME repro on at least plaintext and one void/inline-heavy example
• synthetic helper row only as supporting coverage, not closure
• trace selection before composition start, after cancellation, and after the next typed character

Expected architecture repair:

• composition start records a logical insertion anchor
• uncommitted composition cancellation restores that anchor
• deletion/Esc/click-away paths do not advance the Slate model selection
• follow-up typing inserts at the original logical point

Likely files:

• packages/slate-react/src/editable/composition-state.ts
• packages/slate-react/src/editable/runtime-composition-events.ts
• packages/slate-react/src/editable/selection-reconciler.ts
• packages/slate-browser/src/playwright/ime.ts

Verification target:

PLAYWRIGHT_BASE_URL=http://localhost:3100 playwright test playwright/integration/examples/plaintext.test.ts playwright/integration/examples/richtext.test.ts playwright/integration/examples/mentions.test.ts --project=chromium --grep "IME|composition"

Manual Chrome evidence remains required before any Fixes #6 claim.

Phase 4: Target Ownership, Iframes, Native Controls, And Select-All

Issues: #5, #8, #15.

Add red proof rows first:

• image route Cmd+A after text focus selects editor body text, not only image void UI
• editable void route Cmd+Z after typing into native input removes the newly inserted void if that matches live reference after local proof
• iframe route keeps selected iframe text as toolbar command target when parent toolbar is clicked

Expected architecture repair:

• command target resolution must survive focus moving from the editable to a parent toolbar
• internal native controls must not swallow Slate undo for the transaction that inserted their owning void node when the intended editor target remains active
• select-all must choose the editor body contract, then render void selection UI only when the Slate selection actually targets voids

Likely files:

• packages/slate-react/src/editable/runtime-target-bridge.ts
• packages/slate-react/src/editable/keyboard-input-strategy.ts
• packages/slate-react/src/editable/selection-controller.ts
• packages/slate-react/src/editable/runtime-focus-mouse-events.ts
• playwright/integration/examples/iframe.test.ts
• playwright/integration/examples/editable-voids.test.ts
• playwright/integration/examples/images.test.ts

Verification target:

PLAYWRIGHT_BASE_URL=http://localhost:3100 playwright test playwright/integration/examples/iframe.test.ts playwright/integration/examples/editable-voids.test.ts playwright/integration/examples/images.test.ts --project=chromium

Phase 5: Issue Claim Sync

Only after phases 1-4 are green:

• append exact fork issue dossier closure rows
• update docs/slate-v2/ledgers/issue-coverage-matrix.md only with issues that have exact repro proof
• update docs/slate-v2/references/pr-description.md if a PR claim changes
• run the relevant .tmp/slate-v2 focused checks
• run bun run completion-check in plate-2 only for planning state

Phase 1 Execution Evidence

Executed in /Users/zbeyens/git/slate-v2 on 2026-05-10.

Code owner changes:

• packages/slate/src/transforms-text/insert-text.ts: full-document text replacement now emits one replace_children operation instead of replacing the snapshot without operations, so history can invert the edit.
• packages/slate/src/transforms-text/insert-text.ts: expanded-range replacement uses the post-delete collapsed model selection as the insert target, with point refs as fallback, so formatted-word replacement does not crash when structural deletion nulls both refs.

Proof rows added or confirmed:

• playwright/integration/examples/plaintext.test.ts: #9 middle-line typing undo restores text and caret; #12 partial selection replacement undo restores the original text and selected range.
• playwright/integration/examples/mentions.test.ts: #11 select-all replacement undo restores the original mentions document.
• playwright/integration/examples/hovering-toolbar.test.ts: #14 English replacement over selected formatted bold text does not crash.
• packages/slate-history/test/history-contract.ts: full-document selected text replacement is one undoable replace_children history batch.

Verification:

cd /Users/zbeyens/git/slate-v2
bun --filter slate typecheck
bun --filter slate-history typecheck
bun lint:fix
bun test ./packages/slate-history/test/history-contract.ts
PLAYWRIGHT_BASE_URL=http://localhost:3100 bun playwright test playwright/integration/examples/plaintext.test.ts playwright/integration/examples/mentions.test.ts playwright/integration/examples/hovering-toolbar.test.ts --project=chromium --grep "keyboard undo restores caret after middle-line typing|keyboard undo restores partial selected text replacement|keyboard undo restores select-all replacement content|typing English over selected formatted text does not crash"

Final same-turn results:

• slate typecheck: Exited with code 0
• slate-history typecheck: Exited with code 0
• bun lint:fix: checked 1601 files, fixed 2 files
• history-contract.ts: 24 pass, 0 fail
• focused Chromium Playwright proof: 4 passed

Claim safety:

• udecode/slate#14 has clean local red -> fixed -> green proof in the hovering-toolbar row.
• udecode/slate#11 has green browser proof and package-level history proof after the operation fix. The first local red row mixed the real failure with a Playwright ControlOrMeta/Windows-UA hotkey mismatch, so keep it out of PR-facing Fixes wording until a clean pre-fix artifact is intentionally preserved.
• udecode/slate#9 and udecode/slate#12 are green in the added/existing plaintext proof rows; no code change was needed for those exact rows in this slice.

Phase 2 Execution Evidence

Executed in /Users/zbeyens/git/slate-v2 on 2026-05-10.

Issue-source proof:

• gh issue view 7 -R udecode/slate --json number,title,body,url,comments confirmed the selected image void external paste report.
• gh issue view 10 -R udecode/slate --json number,title,body,url,comments confirmed the selected mention copy/paste and cut crash report.
• gh issue view 13 -R udecode/slate --json number,title,body,url,comments confirmed the paste-html report has only the attached video as repro input.
• The #13 video was downloaded to .tmp/udecode-slate-13.mp4 and translated into a self-copy rendered-content browser row for /examples/paste-html.

Current local red proof:

• udecode/slate#7: a one-off browser probe against http://localhost:3100/examples/images copied the selected image void and found text/html without `` and text/plain containing only newlines.
• Live reference comparison against https://www.slatejs.org/examples/images copied the same kind of selected image void and produced text/html with ``.

Code owner change:

• packages/slate-dom/src/plugin/dom-clipboard-runtime.ts: void selections now clone from before the starting void DOM node through after the ending void DOM node, so spacer-anchored block void selections include visible external HTML content while preserving the internal Slate fragment payload.

Proof rows added or confirmed:

• packages/slate-dom/test/clipboard-boundary.ts: selected block void export carries the image node in the internal fragment and includes visible `` HTML for external clipboard targets.
• playwright/integration/examples/images.test.ts: selected image copy writes text/html with both data-slate-fragment and ``.
• playwright/integration/examples/mentions.test.ts: selected mention copy/paste inserts another mention without runtime errors; selected mention cut removes the mention without runtime errors.
• playwright/integration/examples/paste-html.test.ts: copied rendered Slate content from the paste-html example pastes back as an internal Slate fragment with block structure preserved, before generic HTML import policy matters.

Verification:

cd /Users/zbeyens/git/slate-v2
bun lint:fix
bun test ./packages/slate-dom/test/clipboard-boundary.ts
PLAYWRIGHT_BASE_URL=http://localhost:3100 bun playwright test playwright/integration/examples/images.test.ts playwright/integration/examples/mentions.test.ts playwright/integration/examples/paste-html.test.ts --project=chromium --grep "copies selected image with visible external HTML payload|copies and pastes a selected mention without crashing|cuts a selected mention without crashing|pastes copied rendered Slate content as an internal fragment before HTML import"
bun --filter slate-dom typecheck
bun --filter slate-browser typecheck

Final same-turn results:

• bun lint:fix: checked 1601 files, fixed 1 file
• clipboard-boundary.ts: 25 pass, 0 fail
• focused Chromium Playwright proof: 4 passed
• slate-dom typecheck: Exited with code 0
• slate-browser typecheck: Exited with code 0

Claim safety:

• udecode/slate#7 has current local red -> fixed -> green proof and can be a fork-lane fixes-claimed row.
• udecode/slate#10 is now covered by exact selected mention browser rows and prior inline-void clipboard export work. Do not inflate this to a new same-slice fix claim because the added browser rows were green on current code.
• udecode/slate#13 is now represented by the video-derived self-copy paste-html row. The row is green on current code, so treat it as already-accounted, not a new fix claim.

Phase 3 Execution Evidence

Executed in /Users/zbeyens/git/slate-v2 on 2026-05-10.

Issue-source proof:

• gh issue view 6 -R udecode/slate --json number,title,body,url,comments confirmed the Chinese IME cancellation report and its decisive oracle: after an uncommitted composition is cancelled, the next typed character must insert at the original logical caret position.

Proof row added:

• playwright/integration/examples/rendering-strategy-runtime.test.ts: added keeps canceled IME caret anchored for the next typed character.
• The row places the caret inside text, runs Chromium CDP composition cancellation steps, asserts the model text and selection stay unchanged, types x, and asserts insertion happens at the original caret.

Verification:

cd /Users/zbeyens/git/slate-v2
PLAYWRIGHT_BASE_URL=http://localhost:3100 bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "keeps canceled IME caret anchored for the next typed character"
PLAYWRIGHT_BASE_URL=http://localhost:3100 bun playwright test playwright/integration/examples/rendering-strategy-runtime.test.ts --project=chromium --grep "does not push canceled IME composition onto history|keeps text stable after type-delete-cancel IME composition|keeps canceled IME caret anchored for the next typed character"
bun lint:fix
bun check

Final same-turn results:

• focused Chromium Playwright proof: 1 passed.
• focused Chromium cancellation cluster: 3 passed.
• bun lint:fix: checked 1601 files, no fixes applied.
• bun check: lint, package/site/root typecheck, Bun tests, and Slate React Vitest passed.

Claim safety:

• udecode/slate#6 is already-accounted for the automated Chromium CDP cancellation path. No runtime code changed.
• Do not promote #6 to a PR-facing or manual Chinese IME Fixes claim from this row. The report's raw OS/input-method path still needs manual Chrome/device evidence before exact closure wording.

Phase 4 Execution Evidence

Executed in /Users/zbeyens/git/slate-v2 on 2026-05-10.

Issue-source proof:

• gh issue view 5 -R udecode/slate --json number,title,body,url,comments confirmed the images Cmd+A report.
• gh issue view 8 -R udecode/slate --json number,title,body,url,comments confirmed the editable-void native input undo report.
• gh issue view 15 -R udecode/slate --json number,title,body,url,comments confirmed the iframe toolbar target report.

Current local red proof:

• udecode/slate#5: the new images browser row proved the model selected the full editor text range from text focus, but the first image still rendered selected void chrome during broad text-range select-all.
• udecode/slate#8: the new editable-void browser row proved Cmd+Z from the newly inserted void's native input left two inputs on the page. Desktop Chrome routed the decisive undo through beforeinput historyUndo, so native input undo cleared the input instead of applying Slate history.
• udecode/slate#15: the new iframe browser row was green on current code; the parent toolbar command did not wrongly format the selected iframe text.

Code owner changes:

• packages/slate-react/src/editable/input-controller.ts: beforeinput history events stay model-owned even when the browser target is an internal native control.
• packages/slate-react/src/editable/runtime-before-input-events.ts: native historyUndo and historyRedo from internal controls run through Slate history before the internal-control stop path.
• packages/slate-react/src/editable/runtime-keyboard-events.ts and runtime-root-engine.ts: keydown capture includes a narrow internal-control history fallback for browsers that dispatch undo through keydown instead of beforeinput.
• site/examples/ts/images.tsx: image selection chrome renders only for a direct collapsed image selection, not for broad text-range select-all.
• packages/slate-react/test/kernel-authority-audit-contract.ts: authority inventory was updated for the intentional extra keyboard history render wakeups.

Proof rows added or confirmed:

• playwright/integration/examples/images.test.ts: selects image editor text content from text focus with keyboard select all.
• playwright/integration/examples/editable-voids.test.ts: undo from a new editable void input removes the inserted void block.
• playwright/integration/examples/iframe.test.ts: applies parent toolbar formatting to selected iframe text.
• packages/slate-react/test/editing-kernel-contract.ts: beforeinput history remains model-owned for internal controls.

Verification:

cd /Users/zbeyens/git/slate-v2
bun test ./packages/slate-react/test/editing-kernel-contract.ts
bun --filter slate-react typecheck
PLAYWRIGHT_RETRIES=0 PLAYWRIGHT_BASE_URL=http://localhost:3100 bun playwright playwright/integration/examples/images.test.ts playwright/integration/examples/editable-voids.test.ts playwright/integration/examples/iframe.test.ts --project=chromium --grep "selects image editor text content from text focus with keyboard select all|undo from a new editable void input removes the inserted void block|applies parent toolbar formatting to selected iframe text"
PLAYWRIGHT_RETRIES=0 PLAYWRIGHT_BASE_URL=http://localhost:3100 bun playwright playwright/integration/examples/images.test.ts --project=chromium --grep "deletes selected image|removes an empty paragraph after an image before deleting the image|inserts a paragraph after a clicked selected image on Enter|copies selected image with visible external HTML payload|selects image editor text content from text focus with keyboard select all"
PLAYWRIGHT_RETRIES=0 PLAYWRIGHT_BASE_URL=http://localhost:3100 bun playwright playwright/integration/examples/editable-voids.test.ts --project=chromium --grep "undo from a new editable void input removes the inserted void block|keeps native paste inside editable void input|restores outer editor selection after editing input inside editable void|runs generated internal-control gauntlet without illegal kernel transitions|keeps ArrowLeft inside editable void input native-owned"
bun lint:fix
bun test:vitest test/kernel-authority-audit-contract.test.ts
bun check
PLAYWRIGHT_RETRIES=0 PLAYWRIGHT_BASE_URL=http://localhost:3100 bun playwright playwright/integration/examples/images.test.ts playwright/integration/examples/editable-voids.test.ts playwright/integration/examples/iframe.test.ts --project=chromium --grep "selects image editor text content from text focus with keyboard select all|undo from a new editable void input removes the inserted void block|applies parent toolbar formatting to selected iframe text"

Final same-turn results:

• editing-kernel-contract.ts: 20 pass, 0 fail.
• slate-react typecheck: passed.
• focused Phase 4 Chromium proof: 3 passed.
• images surrounding Chromium proof: 5 passed.
• editable-voids surrounding Chromium proof: 5 passed.
• bun lint:fix: checked 1601 files, fixed 4 files.
• authority audit focused Vitest: 10 passed.
• bun check: lint, package/site/root typecheck, Bun tests, and Slate React Vitest passed.
• final focused Phase 4 Chromium proof after lint/check: 3 passed.

Non-gate observation:

• An accidental broad Chromium integration run passed 313 tests, skipped 5, and failed 5 unrelated rows outside the Phase 4 issue target. The required Phase 4 rows passed inside that broad run and in the clean focused rerun above.

Claim safety:

• udecode/slate#5 has local red -> fixed -> green proof for the exact Cmd+A selected-image visual bug and can be a fork-lane fixes-claimed row.
• udecode/slate#8 has local red -> fixed -> green proof for the editable-void undo bug and can be a fork-lane fixes-claimed row.
• udecode/slate#15 stays already-accounted. The exact browser row is green on current code, so no fix claim is attached to this slice.
• docs/slate-v2/ledgers/issue-coverage-matrix.md and docs/slate-v2/references/pr-description.md are unchanged because these are fork-local udecode/slate issue IDs, not upstream PR auto-close IDs.

ClawSweeper Accounting

No upstream PR-facing issue claim changes. Fork-local accounting is synced in docs/slate-v2/ledgers/fork-issue-dossier.md.

Maintainer Objection Ledger

Applied Skill Notes

• slate-ralplan: applied. This file is the planning/review gate and leaves the full issue set accounted.
• clawsweeper: applied. All 11 open issues are classified with fork-local claim safety; no upstream PR-facing closure claims were written.
• goal workflow: applied through this docs/plans file per repo override.
• learnings-researcher: applied. Prior docs/solutions reinforce that browser IME and clipboard rows need honest browser proof and that Playwright can miss user-visible behavior.
• tdd: applied for the red -> green rows where current behavior failed.
• performance-oracle: skipped for this pass. No performance claim is being made.
• vercel-react-best-practices: skipped for this pass. The current plan is runtime behavior, not React rendering breadth.

Score

Current Ralplan score: 0.93.

Gate: done. Phases 1-4 are complete, fork issue dossier is synced, and upstream PR-facing issue coverage remains unchanged because these are udecode/slate fork issue IDs.

Next Prompt

No next autonomous pass remains for this ralplan. If work continues, use a new ralplan for manual Chinese IME device/browser proof or for the unrelated broad integration failures observed during the accidental full Chromium run.

This ralplan is complete. Start a new focused plan only for manual Chinese IME
proof or unrelated full-suite integration failures.