v1.45.0

Improvements

• Release notes now package-scoped — Release notes are generated with awareness of SuperDoc's scope and architecture. The AI understands your use case (embedding the editor), your bundled internal packages, and writes release notes accordingly — giving you clearer context with each release.
• GitHub Releases page focused on stable versions — Stable releases only appear as GitHub Releases. Prerelease versions continue through Git tags and npm distribution without creating public release entries, keeping your release history clean and relevant.

vscode-v2.17.0

Improvements

• Release notes now VS Code–specific — Release notes are generated with full awareness of the SuperDoc for VS Code extension's scope, audience, and bundled editor. Notes now clarify which changes affect what VS Code users see or can do inside the editor, filtering out irrelevant infrastructure changes.
• GitHub Releases page focused on stable versions — Stable releases now appear as GitHub Releases. Prerelease versions (next channel) continue distributing through npm and Git tags without creating public release entries, keeping your release history clean and relevant.

template-builder-v1.12.2

Fixes

• GitHub release publishing for stable versions only — Prerelease versions no longer create entries on the GitHub Releases page. The GitHub Releases section now reflects your stable release history only, while prerelease distribution continues through Git tags and npm's next dist-tag.

sdk-v1.21.1

Fixes

• Skill name validation at runtime — The SDK now validates skill names against a strict pattern (alphanumeric, underscore, dash) at runtime, catching invalid names early with clear error messages instead of allowing them to fail at install time.
• Skill install options validation — The SDK validates that install options are a plain object, rejecting null, arrays, or primitives with structured errors. Prevents silent failures and malformed installations.
• Table tool documented in Node README — The SDK README now lists superdoc_table in the AI tool catalog, making it discoverable for developers building AI integrations.

sdk-v1.21.0

What's New

• Custom actions authoring kit — Build your own named document operations and expose them to LLMs through superdoc_perform_action. Define actions using defineAction/extendPreset/composePreset (Node) or define_action/extend_preset/compose_preset (Python), with two execution tiers: steps (declarative composition of built-in actions with {{arg}} templating) and run (native function in your process). Cross-language contract enforced — templating semantics, defaults, and receipt shapes are identical between Node and Python.
• One-call toolkit API — Simplified entry point for custom actions: createAgentToolkit/create_agent_toolkit accepts {provider, actions, base?, includeCoreActions?} and returns coherent tools, system prompt, and pre-bound dispatchers in a single call. No preset registration required for simple cases; advanced extendPreset/composePreset paths remain for composed workflows.
• superdoc-custom-actions authoring skill — Distributed via GitHub, guides developers through building custom actions with live-verified examples in JavaScript and Python, covering both execution tiers, async patterns, integration with existing presets, and common pitfalls.

Improvements

• Coherent excludeActions across surfaces — excludeActions/exclude_actions now consistently narrows the action enum, system prompt description, advertised argument properties, and dispatch allowlist together — no separate rebuilding required. Hidden built-in actions are refused at dispatch; custom actions composing hidden built-ins still work (composition is never excluded).
• Smart argument reuse — Custom actions can reuse built-in argument names without conflict — matching type + allowing differing documentation (description, title, examples, comments). Structural differences (type mismatch, incompatible defaults, enum changes) are rejected with clear errors.
• Catalog narrowing with includeCoreActions — composePreset and compose_preset now narrow the tool catalog row — enum, description, and argument properties shrink together with the one implementation, preventing catalog-driven UIs from advertising actions the preset refuses.
• Provider dialect fixes — Vercel dialect now correctly handles flat inputSchema shapes; provider validation enforced upfront in compose paths.

Fixes

• Standalone custom actions preserve action-named arguments — Custom actions with an 'action' argument (as their own provider tools, not via superdoc_perform_action) no longer drop the argument value before defaults and validation.
• Surface exclusions do not bleed into custom steps — Model-facing excludeActions (which hide actions from direct calls) no longer propagate into internal step dispatch, so a curated steps-tier action composing an excluded built-in still succeeds.
• Synthesized perform_action for custom-only presets — When the base preset drops superdoc_perform_action entirely (empty includeCoreActions), active custom actions are now properly advertised and dispatchable via synthesized tool definition.
• Explicit null enum validation — Python enum validation now treats explicit None as a value (matching Node), rejecting {mode: null} with a teaching error instead of passing it through.
• Array immutability for preset surfaces — Node SDK now snapshots the actions array at construction, preventing post-build mutations from desynchronizing advertised tools from dispatchable ones (Python already did this).
• Open schema normalization — Custom actions with open-object schemas (type: object without properties) now normalize to {} and merge correctly instead of throwing during getTools().
• Empty toolkit values preserved — Python SDK no longer drops falsy (0, false, empty string) base preset values when composing toolkits.
• Empty perform-action rows dropped from catalogs — When all core actions are excluded, the perform-action tool is no longer advertised in the catalog if no custom actions are present.

react-v1.16.0

Improvements

• Release notes now package-scoped — Release notes are generated with awareness of this package's scope and audience. GitHub Releases are now limited to stable versions, keeping your release history clean and focused.

mcp-v0.17.1

Fixes

• Skill name validation at runtime — The SDK now validates skill names against a strict pattern at runtime, catching invalid names early with clear error messages instead of allowing them to fail silently.
• Skill install options validation — The SDK validates that install options are a plain object, rejecting null, arrays, or primitives with structured errors to prevent malformed installations.

Improvements

• Table tool now documented — The superdoc_table tool is now listed in the AI tool catalog documentation, making it discoverable for developers building AI integrations.

mcp-v0.17.0

Chores

• 1.21.0 [skip ci]
• 0.22.0 [skip ci]
• 1.44.2 [skip ci]

Bug Fixes

• preserve empty toolkit base values
• drop empty perform-action catalog rows
• publish GitHub releases for stable versions only

Features

• add package-scoped AI release notes to all six public packages (#508)
• custom-actions authoring kit — canonical ActionSpec with steps/run tiers (#410)

fonts-v0.2.0

What's New

• Core preset for AI agents — Replace intent-grouped tools with a deterministic action surface: superdoc_inspect reads document snapshots, superdoc_perform_action applies one of 40 named, validated edits and returns receipts with verification. Pass preset: 'core' to createAgentToolkit() (Node.js and Python) or specify it in CLI preset commands. Two tools guarantee consistency across your agent loop; mixing presets between hand-assembled calls is impossible.
• trackChanges.decide operation — Accept or reject tracked changes by ID, by text range, by logical anchor range, or all at once. Range targets can partially resolve a change, splitting surviving content into new fragments with stable IDs. Optional move-pairing assertions verify change intent before deciding. Supports optional side parameter to resolve only one half of paired replacements.

Improvements

• Agent toolkit ensures preset consistency — Node.js and Python SDKs now offer createAgentToolkit() that returns tools, system prompt, and dispatcher all bound to the same preset. Eliminates the most common integration mistake: mixing preset tool definitions with legacy prompts.
• CLI preset commands — Query and dispatch agent presets from the command line: preset list, preset get-catalog, preset get-tools, preset get-system-prompt, and preset dispatch for running tool calls against live documents with automatic session tracking.
• Hyperlink handling improvements — Better round-trip fidelity for hyperlinks in imported and exported documents, including improved handling of complex hyperlink attributes and relationships.
• Table cell rendering improvements — Refined table cell border logic and cell content layout for more accurate visual representation of complex table structures with nested content and conditional styling.
• Document property cascade improvements — Enhanced paragraph and list properties resolution to better respect OOXML style cascades and conditional formatting rules.
• GitHub releases stable-only — Release pages now focus on shipped stable versions; prerelease versions skip GitHub releases but maintain Git tags and npm next distribution. Linear prerelease breadcrumbs link to Git tags instead of missing release pages.

Fixes

• Tracked change range decisions now support partial resolution — Range-based decisions that span less than the full tracked change now properly split the change into fragments instead of accepting or rejecting the entire logical change.
• Comment anchor remapping on tracked change decisions — Comments anchored within partially decided tracked changes are now correctly remapped to surviving content, and cascade-deleted comment IDs are reported in receipts.

esign-v2.7.3

Bug Fixes

• publish GitHub releases for stable versions only

Changes

• Merge pull request #101 from superdoc/artem/SD-3452
• 🔄 Sync stable → main