sdk-v1.20.0

What's New

Core preset for AI agents — Ships superdoc_inspect and superdoc_perform_action with 40 LLM-safe actions: change text, rewrite blocks, add comments, move/delete tables, attach numbering, reply to comments, format text, and more.

Tracked paragraph-property changes — Agents and users can now track numbering and alignment changes (w:pPrChange). Full round-trip: track it, export to Word, import back, accept/reject in the review UI.

Comment threading — reply_to_comment action with proper parentId wiring so threads survive import/export.

Citation hyperlinks with tracked changes — Hyperlinks now preserve w:ins/w:del marks through DOCX export and re-import.

MCP core preset support — Serve the core preset tools over Model Context Protocol with MCP_PRESET=core.

createAgentToolkit — Single options object returns {tools, systemPrompt, dispatch} with preset + exclusions applied to all three; no more forgetting to pass excludeActions to every function.

Provider-native tool formats — getTools supports 'vercel' (flat), 'openai' (Chat-Completions-nested), and 'anthropic' (with cache_control).

Improvements

Large-document performance — 50s → 16s on 38-page redlines. Headless view-only work skips whole-doc linked-style CSS decorations; blocks.list returns visible text (skips tracked-deleted runs) so re-editing resolves offsets correctly.

Tracked-change decision interface — Side-targeted reject: pass side: 'inserted' | 'deleted' to trackChanges.decide and resolve only half a replacement.

System prompt narrowing — excludeActions parameter shrinks both the tool enum and the prompt documentation so the model sees no phantom actions.

Receipt token hygiene — Per-item lists capped at 8 entries; full counts in executedOperationCount / selectedTargetCount. Stops 200-paragraph font changes from costing thousands of prompt tokens.

Python SDK parity — Core preset, actions, dispatch with exclusion guards, create_agent_toolkit, full ActionArgs types.

Tracked-change accept/reject now syncs numberingProperties and listRendering — Rejecting a tracked numbering change restores both the block's paragraph properties and the document's top-level list state.

Fixes

Cell shading on export — tables.setShading({ color: null }) now deletes background attr on both table-scoped and cell-scoped paths.

Tracked pPrChange edge cases — One-sided replacement survivors downgrade to plain insert/delete. pPrChange inside kept tracked tables resolves via id, not through mark-based child planner. Tracked numbering attach guards against blank authors.

Border rendering — Thick borders paint at authored width (no 2x multiplier).

Native CLI binary prompt assets — Bun-compiled CLI binaries now embed system-prompt.md and serve get-system-prompt without TOOLS_ASSET_NOT_FOUND.

Hyperlink edge case — Final-doc export flattens tracked pPrChange so .docx carries only accepted numbering; accepting a tracked table by id cascades to resolve contained pPr changes.

Prompt asset errors — readPromptFile now distinguishes ENOENT (missing asset) from permission/IO failures.

CLI tracked mode transport — attach_numbering, add_list_items, lists.setType in changeMode:'tracked' now pass the flag in both input and options dialects so tracked changes survive CLI transport.

move_range safety — Refuses ranges containing tables, lists, or images before mutating.

reply_to_comment contract fix — Sends only {text, parentId, parentCommentId} with no target so threads inherit parent's anchor per spec; both SDK transport and in-process hosts now work.

Synthetic tracked-change rows no longer leak as empty comments on export — comments.list() returns synthetic rows for the sidebar, but exporting them with exportDocx({ comments }) no longer fabricates spurious w:comment elements.

mcp-v0.16.0

What's New

Core preset for AI agents — Ships superdoc_inspect and superdoc_perform_action with 40 LLM-safe actions: insert/replace/delete text, add comments, create tables, format content, accept/reject changes, and more. All verbs mapped to document operations.

Tracked paragraph-property changes — Numbering and alignment changes now round-trip as w:pPrChange. Track it, export to Word, re-import, and accept or reject in the review UI.

Comment threading — reply_to_comment action with proper parentId wiring so conversations survive export/import.

Citation hyperlinks with tracked changes — Hyperlinks now preserve w:ins/w:del marks through DOCX export and re-import.

MCP core preset support — Serve the core preset tools over Model Context Protocol with MCP_PRESET=core.

createAgentToolkit — Single options object returns {tools, systemPrompt, dispatch} with preset and exclusions applied to all three — no more forgetting to pass excludeActions to every function.

Provider-native tool formats — getTools supports 'vercel' (flat), 'openai' (Chat-Completions-nested), and 'anthropic' (with cache_control).

Improvements

Large-document performance — 50s → 16s on 38-page redlines. Headless editor skips whole-doc linked-style CSS. blocks.list returns visible text (skips tracked-deleted runs) so re-editing resolves offsets correctly.

Tracked-change decision interface — Side-targeted reject: pass side: 'inserted' | 'deleted' to trackChanges.decide and resolve only half a replacement.

System prompt narrowing — excludeActions parameter shrinks both tool enum and prompt documentation — model sees no phantom actions.

Receipt token hygiene — Per-item lists capped at 8 entries; full counts in executedOperationCount and selectedTargetCount. Stops 200-paragraph font changes from costing thousands of prompt tokens.

Python SDK parity — Core preset, 40 actions, dispatch with exclusion guards, create_agent_toolkit, full ActionArgs types.

Fixes

Cell shading on export — tables.setShading({ color: null }) now deletes background attr on both table-scoped and cell-scoped paths.

Tracked pPrChange edge cases — One-sided replacement survivors downgrade to plain insert/delete. pPrChange inside kept tracked tables resolves via id, not mark-based planner. Tracked numbering attach guards against blank authors.

Border rendering — Thick borders paint at authored width (no 2x multiplier).

Native CLI binary prompt assets — Bun-compiled binaries now embed system-prompt.md and serve get-system-prompt without TOOLS_ASSET_NOT_FOUND.

Hyperlink round-trip — Final-doc export flattens tracked pPrChange so .docx carries only accepted numbering. Accepting a tracked table by id cascades to resolve contained pPr changes.

Prompt asset errors — readPromptFile now distinguishes ENOENT (missing asset) from permission/IO failures.

CLI tracked mode transport — attach_numbering, add_list_items, lists.setType in changeMode:'tracked' now pass the flag in both input and options dialects so tracked changes survive transport.

move_range safety — Refuses ranges containing tables, lists, or images before mutating.

reply_to_comment contract — Sends only {text, parentId, parentCommentId} with no target so threads inherit parent's anchor per spec.

Comment export cleanup — Synthetic tracked-change rows for the sidebar no longer fabricate spurious w:comment elements on export.

cli-v0.21.0

What's New

Core preset for AI agents — Ships two statically validated tools (superdoc_inspect, superdoc_perform_action) with 40 LLM-safe actions: change text, rewrite blocks, add comments, move/delete tables, attach numbering, reply to comments, format text, and more.

Tracked paragraph-property changes — Agents and users can now track numbering and alignment changes (w:pPrChange). Full round-trip: track it, export to Word, import back, accept/reject in the review UI.

Comment threading — reply_to_comment action and proper parentId wiring in comments.create so threads survive import/export.

Citation hyperlinks with tracked changes — Hyperlinks now preserve w:ins/w:del marks through DOCX export and re-import (fixed structural insertion of tracked citations).

MCP core preset support — Serve the core preset tools over Model Context Protocol with MCP_PRESET=core.

Improvements

Large-document performance — 50s → 16s on headless 38-page redlines. Skips whole-doc linked-style CSS decorations when the editor runs view-only, and blocks.list now returns visible text (skips tracked-deleted runs) so re-editing resolves offsets correctly.

Tracked-change decision interface — Side-targeted reject: pass side: 'inserted' | 'deleted' to trackChanges.decide and resolve only half a replacement. Survives one-sided deletions correctly.

System prompt narrowing — excludeActions parameter shrinks both the tool enum and the prompt documentation (no phantom actions, no hallucination bait). getSystemPrompt and getTools stay coherent by construction.

createAgentToolkit — Single options object returns {tools, systemPrompt, dispatch} with preset + exclusions applied to all three. No more forgetting to pass excludeActions to every function.

Receipt token hygiene — Per-item lists (executedOperations, selectedTargets) capped at 8 entries; full counts in executedOperationCount / selectedTargetCount. Stops 200-paragraph font changes from costing thousands of prompt tokens.

Python SDK parity — Core preset, actions, dispatch with exclusion guards, create_agent_toolkit, full ActionArgs types.

Provider-native tool formats — getTools supports 'vercel' (flat {name, description, inputSchema}), 'openai' (Chat-Completions-nested), 'anthropic' ({name, description, input_schema} with cache_control).

Fixes

Cell shading on export — tables.setShading({ color: null }) now deletes background attr on both table-scoped and cell-scoped clear paths; cells stop rendering as shaded after a cleared shade is re-exported.

Tracked pPrChange edge cases — One-sided replacement survivors downgrade to plain insert/delete. pPrChange inside kept tracked tables resolves via id, not through mark-based child planner. Tracked numbering attach guards against blank authors. Synthetic tracked-change rows no longer leak as empty comments on export. pPrChange no longer positional in review graph (no comment detach, no text-range capture).

Border rendering — Thick borders paint at authored width (no 2x multiplier); test aligned with SD-3028 spec.

Native CLI binary prompt assets — Bun-compiled CLI binaries embed system-prompt.md and now serve get-system-prompt without TOOLS_ASSET_NOT_FOUND.

Hyperlink edge case — Final-doc export flattens tracked pPrChange so .docx carries only accepted numbering; accepting a tracked table by id cascades to resolve contained pPr changes.

Prompt asset errors — readPromptFile now distinguishes ENOENT (missing asset) from permission/IO failures (TOOLS_ASSET_UNREADABLE with underlying cause).

CLI tracked mode transport — attach_numbering, add_list_items, lists.setType in changeMode:'tracked' now pass the flag in both input and options dialects so tracked changes survive CLI transport.

move_range safety — Refuses ranges containing tables, lists, or images before mutating (teaches the model the constraint and prevents silent data loss).

reply_to_comment contract fix — Sends only {text, parentId, parentCommentId} with no target (threads inherit parent's anchor per spec); both SDK transport and in-process hosts now work.

sdk-v1.16.0-next.51

1.16.0-next.51 (2026-07-05)

Bug Fixes

• ci: install SDK Python test deps in release workflows (b4ec138), closes superdoc-dev/superdoc#3798
• cli: build SDK before native release artifacts (d3b1f0c), closes superdoc-dev/superdoc#3795

mcp-v0.12.0-next.53

0.12.0-next.53 (2026-07-05)

Bug Fixes

• ci: install SDK Python test deps in release workflows (b4ec138), closes superdoc-dev/superdoc#3798

mcp-v0.12.0-next.52

0.12.0-next.52 (2026-07-05)

Bug Fixes

• cli: build SDK before native release artifacts (d3b1f0c), closes superdoc-dev/superdoc#3795

cli-v0.17.0-next.48

0.17.0-next.48 (2026-07-05)

Bug Fixes

• cli: build SDK before native release artifacts (d3b1f0c), closes superdoc-dev/superdoc#3795

v1.44.0

Based on my analysis of the commits and codebase changes, here are the release notes:

What's New

• Side-targeted tracked change decisions — Accept or reject only the inserted or deleted half of a paired replacement via trackChanges.decide() with the side parameter ('inserted'/'deleted'), leaving the other half as a standalone pending change.
• Tracked paragraph property changes in review API — Tracked numbering and alignment changes (w:pPrChange) now surface in trackChanges.list() and can be accepted/rejected via trackChanges.decide().
• List numbering metadata — blocks.list returns computed numbering (marker, path, kind) for numbered list items and numbered headings, plus a paragraphNumbering reference field so agents can read legal clause numbers.
• Paragraph indentation metadata — blocks.list now exposes paragraph indentation (left, right, firstLine, hanging in twips) for layout-aware document analysis.
• Visible text model for blocks.list — Text, length, preview, and ref now use the visible text model (tracked deletions excluded), matching how edit refs resolve and eliminating offset drift.
• Tracked list attach — lists.attach() now supports tracked mode for list property changes.
• Citation hyperlinks round-trip — Tracked citation hyperlinks export and import correctly through DOCX with full revision history preserved.
• LLM tools core preset — CLI and MCP expose the execute-code command for running JavaScript directly against documents in headless mode.

Fixes

• Synthetic tracked-change rows no longer leak as empty comments in comments.list() export.
• Cell-scoped tablesClearShading() now removes the background attribute correctly.
• Formatting samples from blocks.list skip deleted runs — no longer reports styling from pending deletions.
• Redlined blocks with tracked deletions at the start no longer throw "offset out of range" when edited via ref.
• Headless mode skips whole-document linked-style decorations for ~3x faster large-document load (38-page redline: 50s → 16s).
• Merge conflict markers in painter-dom border-utils resolved; build no longer fails after main-into-stable merge.

vscode-v2.16.0

What's New

• Tracked-change side-targeted decisions — trackChanges.decide() now accepts a side selector to reject only the inserted or deleted half of a replacement.
• Tracked paragraph-property changes — Tracked numbering and alignment changes (w:pPrChange) now appear in trackChanges.list() and can be accepted or rejected via trackChanges.decide().
• Tracked list operations — lists.attach() now supports tracked changes.
• Comment export — Comments and comment replies now export to DOCX, including comments on tracked changes.
• List numbering metadata — blocks.list() now returns numbering and indent fields so agents can read computed list markers and legal clause numbers.
• Visible text model — blocks.list() returns the visible text model (skips tracked deletions), matching how the plan engine applies edits. Eliminates text offset drift on redlined documents.
• CLI preset commands — New commands to introspect, configure, and dispatch preset operations: preset list, preset get-catalog, preset get-tools, preset get-system-prompt, preset get-mcp-prompt, preset dispatch.
• CLI execute-code command — New command to execute arbitrary operations on a document.
• LLM tools core preset — New preset providing AI-agent-optimized operations for document manipulation.

Improvements

• Large-document performance in headless mode — linked-style decorations are skipped when the editor runs view-only (~50s → ~16s on a 38-page redline).
• blocks.list() length, preview, and block references now use the visible text model, eliminating stale offset calculations on edited blocks.
• Tracked change decision validation tightened — side selector on id-target changes fails closed if the targeted half was already resolved.
• Tracked paragraph-property changes now sync both paragraph and top-level numbering properties on reject.
• Block formatting sampling now draws from visible runs only, skipping tracked deletions.
• Comment export filters synthetic tracked-change rows from the comments list.

Fixes

• Citation hyperlinks now export and import correctly as tracked changes.
• Merge conflict markers in border-utils resolved.
• Cell shading clear operations now properly remove the background attribute on cell-scoped clears.
• Border rendering aligned with SD-3028 authored-width specification.
• Tracked paragraph-property changes export with Word-safe decimal IDs instead of UUIDs.
• One-sided replacement survivors now downgrade to plain insertions or deletions instead of failing.
• Tracked paragraph-property changes inside kept tracked tables now resolve via ID-based decisions instead of the mark-based child planner.
• Tracked changes without a user are now guarded to prevent blank author stamps.

react-v1.15.0

Based on my analysis of the commits and changed files, here are the release notes:

What's New

• Agent tools for document automation — superdoc_inspect and superdoc_perform_action tools expose the full Document API to language models. Run headless in Node.js; bring your own LLM for redlining, template workflows, and document automation.
• Comment API and export — Full CRUD on comments: create, edit, delete, resolve/reopen, move between ranges. Comments export to DOCX, including comments anchored to tracked changes. Anchor comments to tracked-change IDs directly without calculating text ranges.
• Side-targeted tracked-change decisions — Accept or reject just the inserted or deleted branch of a replacement, leaving the other half as a standalone change. Decide by id or by range.
• Tracked paragraph-property changes in the review API — Tracked numbering and alignment changes (w:pPrChange) now surface in trackChanges.list and can be accepted/rejected via trackChanges.decide, matching tracked text behavior.
• List and paragraph numbering metadata — blocks.list returns computed numbering (marker, path, kind) for clause-numbered headings and list items, plus paragraph indentation for layout-aware agents.
• Agent presets for OpenAI, Anthropic, Vercel, and generic providers — Tool catalog and system prompt selection by preset; dispatch tool calls through the preset's validated schema.
• CLI commands for agent presets — preset list, preset get-catalog, preset get-tools, preset get-system-prompt, preset get-mcp-prompt, preset dispatch.
• Tracked citation hyperlinks — Export and import tracked hyperlinks with revision marks preserved, so Word round-trips linked tracked text correctly for reviewable citations.

Improvements

• Large-document performance — Skip linked-style decorations in headless mode; ~50s → ~16s on a 38-page redline.
• Visible text model in blocks.list — Text, length, ref, and formatting now use visible text (skipping tracked-deleted runs), so agents resolve offsets against the same text the plan engine applies against.
• Word-safe revision IDs — Tracked changes export with decimal IDs instead of UUIDs, so Word accepts them without repair.
• Better tracked-change edge cases — Handle one-sided replacement survivors, pPrChange inside kept tracked tables, and no-user guards for tracked operations.

Bug Fixes

• Resolve merge conflict markers in border-utils that broke the build.
• Citation hyperlinks now export with correct revision mark wrapping (w:hyperlink containing w:ins/w:del).
• Reject a tracked table by id now cascades to contained paragraph-property changes.
• Cell background shading now clears on cell-scoped tables.setShading({ color: null }).
• Tracked changes no longer leak as spurious empty comments in export.
• Formatted text from styled tracked-deleted runs no longer bleeds into blocks.list formatting.