47 commits from the align branch merged onto the force-updated remote
base (which had replayed an earlier snapshot of the same work plus new
overlay/pointer features and CI fixes). Conflict resolution: kept the
newer align side for the generation pipeline (orchestrator, mcp, skills,
design tools), kept the base side for the chat-panel test semantics and
graceful overlay teardown, fused both in sub_agent_session (design-turn
thinking policy + graceful epoch finish), and dropped the files each
side had deleted (legacy concurrent/dashboard paths, retired TS skills).
Deduped two identical replayed hunks (export.rs, chat_session_tests.rs).
Known issue carried over: provider_probe_host::landed_connected_outcome_
without_models_is_failure fails on a host with a live provider config
(env-sensitive test, both sides byte-identical there; green on CI).
Built-in design generation now runs as an agentic MCP tool-loop (reusing the
agent-rs BuiltInProvider), gated behind OPENPENCIL_DESIGN_AGENT_LOOP / the
Settings experimental toggle; the orchestrator stays the default.
- design-agent system prompt + in-process design toolset (parity-locked with
the MCP surface) + flag-gated Intent::Design routing
- spawn_agents execution as sequential sub-loops + live creation-mode badges
(per-agent glow + 'N/M designing...' header)
- new MCP tools: get_guidelines, ToolSearch, get_screenshot, get_editor_state,
export_nodes, spawn_agents; style-guide local audit
- #27 AI panel restyle: rounded tool cards + green check-rings, gray user
bubbles, model-pill bottom toolbar, header, empty-state pills, the
PARALLEL AGENTS (agent_team_size) 1x-6x chip dropdown
- multi-chat tabs: ChatSessions model (Deref-to-active) + tab row UI
(switch / close / + / Cmd+T) with each run bound to its tab
Large checkpoint commit spanning the working tree (Rust shell crates).
The product is now Rust (crates/) + the Zig agent runtime
(packages/agent-native) + the wasm-backed web SDK (packages/op-web-sdk*).
Delete the retired TypeScript editor/desktop/CLI (apps/*), the pen-*
packages, and their orphaned TS tooling, Dockerfile, and root tsconfig.
Rust-side hooks into the deleted TS are rehomed so the workspace still
builds, tests, and releases green:
- op-mcp: vendor the 10 element-tool-defs shards it include_str!'d from
pen-mcp into crates/op-mcp/assets/element-tool-defs/ (git renames), and
repoint the compile-time includes + the runtime parity test's read_dir.
- op-orchestrator: reword the parity panic that named the retired
dump-planner-golden.ts generator (goldens stay as frozen baselines).
- CI: drop the TS-derive golden-drift + planner-parity oracle jobs and the
TS path filters from rust-check.yml; repoint the cargo-bundle icon in
rust-release.yml to crates/op-host-desktop/assets/icon.png.
- Release (Windows): recover the deleted apps/desktop/build/icon.ico into
crates/op-host-desktop/assets/icon.ico and repoint the NSIS installer
(rust-release.yml /DICON_FILE + package-windows.nsi ICON_FILE fallback +
its example invocation), so the Windows release build still finds it.
- Remove the dead .githooks/post-commit hook — it rebuilt the deleted
out/mcp-server.cjs via the removed mcp:compile script and only watched
now-deleted pen-* paths; the Rust MCP is built by cargo.
- Recover apps/web/public/logo-discord.svg to screenshot/ and repoint the
15 READMEs' logo/link paths.
package.json is now Rust-first (dev/build/test -> cargo + serve-web) and
docs (CLAUDE.md/AGENTS.md/packages+crates CLAUDE.md) describe the Rust
product. The retired TS remains recoverable via git (last tag v0.7.5).
Verified: cargo check --workspace green; cargo test -p op-mcp 353 passed;
cargo test -p op-orchestrator 848 passed; cargo test --workspace --no-run
compiles all targets; rust-release.yml parses and references no deleted
build assets.
Pre-commit fmt hook bypassed: its cargo fmt --check is pre-red from a
repo-wide stable-vs-nightly rustfmt config drift affecting unrelated files;
the .rs files touched here produce no fmt diff.
* feat(boolean-operations): implement boolean operations in the editor
- Added a new BooleanToolbar component for union, subtract, and intersect operations.
- Integrated boolean operations into the layer context menu and keyboard shortcuts.
- Enhanced the editor layout to include the boolean toolbar for improved user interaction.
- Updated internationalization support with new translation keys for boolean operations.
- Bumped version to 0.3.0 to reflect the addition of these features.
* refactor(editor): update editor layout and panels for improved functionality
- Replaced the PropertyPanel with a new RightPanel that includes both Property and Code panels.
- Removed the CodePanel from the main editor layout and integrated it into the RightPanel.
- Updated keyboard shortcuts to switch the right panel to the code tab.
- Enhanced the LayerPanel with a resizable width feature for better user experience.
- Added internationalization support for new right panel labels and code panel features.
- Introduced new code generation capabilities for various frameworks in the CodePanel.
- Improved overall layout structure for better responsiveness and usability.
* feat(electron): implement .op file association and enhance file handling
- Added support for .op file association in electron-builder, allowing OpenPencil documents to be opened directly from the file system.
- Implemented IPC handlers for opening and reading .op files, ensuring proper loading of document content.
- Enhanced the main process to handle file opening events on macOS and single-instance locking on Windows/Linux.
- Updated the renderer to listen for file open events and load documents accordingly.
- Improved README to reflect new file association feature.
* fix(canvas): improve layout accuracy for AI-generated designs
- Unify lineHeight default via canonical defaultLineHeight() function
- Unify text measurement by removing duplicate estimators in generation-utils
- Fix optical centering formula to scale proportionally with fontSize
- Round layout positions to whole pixels to prevent sub-pixel artifacts
- Recursively sanitize nested x/y in streaming layout containers
- Fix input trailing icon alignment using fill_container instead of space_between
* feat(canvas): right-align agent badge and add breathing glow border
- Agent badge now right-aligned to frame's right edge instead of after label
- Added breathing glow border around agent-owned frames during generation
- Glow border uses same color and lifecycle as the agent badge
- Removed unused BADGE_GAP constant and useDocumentStore import
* feat(code-panel): enhance tab scrolling functionality and add scrollbar utility
- Introduced left and right scroll buttons for tab navigation in the CodePanel, improving user experience for navigating long tab lists.
- Added a custom utility to hide scrollbars for a cleaner interface.
- Updated styles for better responsiveness and usability in the CodePanel layout.
* fix(docs): update Discord invite links in multiple README files
- Replaced outdated Discord invite links with the new link across all language-specific README files.
- Ensured consistency in the documentation for community engagement.
* feat(code-panel): enhance system prompt for responsive design
- Updated the ENHANCE_SYSTEM_PROMPT to emphasize the importance of responsive design in code rewriting.
- Added detailed guidelines for converting fixed pixel widths to relative units and using responsive Tailwind breakpoints.
- Ensured that the output remains visually faithful on desktop while adapting gracefully across screen sizes.
* feat(docs): add WeChat group information to README.zh.md and include group image
- Introduced a new section in the Chinese README to provide details about the WeChat group for community engagement.
- Added an image representing the WeChat group for better visibility and user interaction.
* feat(electron): enhance theme management and title bar overlay for Windows/Linux
- Updated the `setTheme` method in the Electron API to accept custom colors for the title bar overlay, improving theme synchronization across platforms.
- Adjusted title bar overlay colors for Windows and Linux to ensure proper visibility and aesthetics.
- Enhanced the top bar component to read computed CSS colors and apply them dynamically, ensuring a consistent user interface.
- Improved handling of theme changes in the application to support background and foreground color customization.
* fix(screenshot): update screenshot image for improved clarity and quality
* fix(docs): update WeChat group image path in README.zh.md for consistency
* fix(ai): fix post-generation validation pipeline and text centering
- Fix Agent SDK validation: save temp screenshots inside project dir
(.openpencil-tmp/) so Claude Code plan mode can read them, instead
of /tmp/ which is outside the project sandbox
- Enrich validation tree dump with fill colors, stroke, fontSize,
fontWeight, textAlign, cornerRadius, opacity for comprehensive
visual analysis
- Add multi-round validation with quality scoring (threshold 8/10),
500ms stabilization delay between rounds
- Add detailed debug logging to applyValidationFixes showing which
nodes were found/skipped and property changes
- Fix canvas sync needsTextbox check to also account for textAlign
(matching isFixedWidthText in factory), preventing IText↔Textbox
thrashing on every sync tick
- Auto-center text in vertical+center layouts by expanding to full
container width and injecting textAlign:'center'
- Force Textbox for non-left-aligned text so textAlign is respected
(IText ignores width and computes its own)
* fix(canvas): use precise text width estimation for fit-content layout
Remove the 14% safety factor from text width estimation when computing
fit-content/natural-width text dimensions. IText auto-computes its own
width and ignores our setting, so the safety margin only inflated the
layout allocation, making text appear left-shifted within its container.
* fix(canvas): center fit-content text in horizontal layouts
For text nodes with fit-content width in horizontal layouts, set
textAlign:'center' to compensate for width estimation inaccuracy.
The estimated box is typically wider than the actual rendered text,
causing left-aligned text to appear visually shifted. Centering
distributes the estimation error evenly on both sides.
* feat(ai): show validation details in checklist panel
- Accumulate validation log (screenshot, analysis, fixes) instead of
overwriting status messages, so the full process is visible
- Preserve step thinking content in buildFinalStepTags (was discarded)
- Add details field to pipeline items and render in checklist UI
- Each validation step now shows: screenshot captured, issues found,
quality score, fixes applied
* feat(ai): add visual reference pipeline types and integration hooks
- Add DesignSystem and VisualReference types to ai-types
- Add 'visual-ref' mode to AIDesignRequest and SubTask.htmlReference
- Detect visual-ref candidates in chat handlers (landing pages, websites)
- Wire visual-ref mode in design-generator and orchestrator
- Inject HTML reference snippets into sub-agent prompts
* feat(ai): add modular design principles for sub-agent context
- Add design-principles module with topic files: color, typography,
spacing, composition, components
- Selectively load relevant principles based on prompt content
- Inject design principles into sub-agent system prompts
* feat(ai): implement visual reference pipeline
- Add design-system-generator: generates color/typography/spacing tokens
- Add design-code-generator: generates HTML/CSS from design system
- Add html-renderer: renders HTML to screenshot via html2canvas
- Add visual-ref-orchestrator: coordinates the full pipeline
(design system → HTML code → screenshot → enrich subtasks)
- Add html2canvas dependency for client-side HTML rendering
* feat(mcp): default filePath to live canvas and fix cross-platform issues
- Default all MCP tool filePath to live://canvas when omitted, so tools
operate on the real-time canvas instead of stale files
- Remove filePath from required params in all tool schemas (21 interfaces)
- Fix mcp-server-manager.ts using process.cwd() which fails in Electron
production on Linux — now checks ELECTRON_RESOURCES_PATH first
- Fix stopMcpHttpServer using SIGTERM on Windows — use taskkill instead
- Force new children reference in applyExternalDocument to ensure canvas
sync subscriber always detects MCP-pushed document updates
* feat(mcp): enhance design prompt with semantic roles, CJK typography, and layout rules
Add comprehensive design knowledge to MCP design prompt for better
AI-generated designs: design type detection (mobile vs desktop), full
semantic role reference with context-aware defaults, CJK typography
rules, expanded text/layout/form guidelines, and detailed post-processing
documentation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(ai): implement intent classification for chat handlers
- Replace hardcoded keyword matching with a lightweight LLM call to classify user intent in chat messages.
- Introduce a new function `classifyIntent` to determine if the request is for design generation or conversation.
- Update design request handling in `useChatHandlers` to utilize the new classification method.
- Enhance design prompt documentation to reflect changes in design type detection based on intent rather than keywords.
* fix(ai): handle string qualityScore in validation response parsing
The LLM sometimes returns qualityScore as a string (e.g. "8" instead of 8),
causing it to fall through to 0. Also hide misleading "quality: 0/10" display
when the score couldn't be determined, and log raw response for debugging.
* fix(ai): increase validation timeout to 90s and fix quality score parsing
Agent SDK validation requires spawning a process, reading the image, and
analyzing it — 30s was consistently timing out. Also handle string
qualityScore values from LLM responses and hide misleading 0/10 display.
* fix(ai): fix validation timeout and response parsing
- Increase validation timeout from 30s to 180s (Agent SDK needs time
for subprocess spawn + OAuth auth + multi-turn image reading)
- Strip <tool_use> XML blocks from Agent SDK response before extracting
JSON — the tool call XML was confusing the regex, causing qualityScore
to parse as 0 despite valid JSON being present
- Handle string qualityScore values and hide misleading "quality: 0/10"
- Revert unnecessary direct API key approach for validation
* fix(ai): prevent node ID collisions between generations
When generating new content on a canvas with existing nodes, AI-generated
IDs (e.g. brand-spacer) would collide with previous generations. Now
captures pre-existing node IDs at generation start and checks against
them during upsert sanitization. Remapped IDs are tracked in
generationRemappedIds so progressive streaming updates can still find
their nodes.
* fix(ai): require styleGuide in orchestrator plan and fix validation detail icons
- Add fallback default styleGuide when orchestrator LLM omits it
- Strengthen prompt to mark styleGuide as REQUIRED
- Replace emoji icons in validation details with [done]/[pending]/[error]
markers for consistent styling with the checklist design system
* feat(server): add port file plugin for server instance discovery
- Introduce a new Nitro plugin that writes a port file on server startup to allow the MCP server to discover the running instance, whether it's a development server or Electron.
- Implement error handling in the Electron main process for writing the port file, logging any failures.
- Update Vite configuration to include additional external dependencies in the rollup configuration.
* feat(electron): implement IPC for retrieving pending file paths
- Added a new IPC handler `file:getPending` to retrieve and clear the pending file path when the React app mounts.
- Updated the Electron API to include `getPendingFile` for renderer access.
- Enhanced the `useElectronMenu` hook to load any pending file on application startup.
- Updated UI components to reflect changes in file handling and improved user experience.
* fix(panels): replace emoji icons with styled icons in validation checklist
- Parse [done]/[pending]/[error] prefixes in detail lines and render as
styled circle icons matching the parent checklist design system
- Replace remaining emoji markers in design-validation.ts with text prefixes
- Fix isApplied detection to recognize new [done] Applied marker
* refactor(electron): update settings path to use platform-standard app data directory
- Changed the settings file path to utilize Electron's user data directory for better cross-platform compatibility.
- Updated the settings writing function to ensure the user data directory is created if it doesn't exist.
- Added comments to clarify the storage location for different operating systems.
- Implemented a fixed partition for localStorage/cookies to maintain data across server port changes.
* feat(ai): enhance validation with pre-checks, structural fixes, and border detection
- Add design-pre-validation.ts: pure code checks before LLM validation
- Invisible container detection (same fill as parent → auto-add border)
- Sibling consistency (majority-rule for height/cornerRadius)
- Add structural fixes to validation: addChild/removeNode operations
- Icon injection via lookupIconByName with server fallback
- autoFixParentLayout with child count guard to prevent layout breakage
- Add strokeColor/strokeWidth to safe fix properties for border fixes
- Simplify intent classification: all design requests use visual-ref pipeline
- Fix checklist: "Found N issues" now shows [done] instead of [pending]
- Fix qualityScore: only update when > 0 to preserve valid round scores
* fix(ai): cherry-pick safe validation improvements, drop aggressive pre-checks
Keep: stroke tree dump bug fix (object not array), qualityScore=0 false
positive detection, fit_content→fixed safety guard, empty path removal,
type-specific sibling consistency, repeated fix filtering, screenshot
extraction to design-screenshot.ts.
Drop: detectForcedFixedHeight (destroyed input/button heights),
MAX_VALIDATION_ROUNDS 5 (too many rounds), removal of quality threshold
early stop, section regeneration phase.
---------
Co-authored-by: Fini <fini.yang@gmail.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* feat(editor): add github releases auto update flow
Enable desktop auto-update checks and install prompt so packaged apps can update from published GitHub Releases.
* fix(canvas): keep frame labels aligned during zoom
Draw labels in viewport space with zoom-aware positioning so frame titles stay anchored while panning and zooming.
* fix(ai): disable chat interactions when no models connected
Prevent sending and shortcut actions when provider models are unavailable, and show a concise model picker hint to avoid false-available UI states.
* docs: refresh readme community and demo section
Embed the product demo video inline, add a styled Discord community entry, and update media assets for clearer README presentation.
* chore: update nitro external dependency settings
Align SSR bundling with current runtime usage by removing the unnecessary @opencode-ai external entries.
* feat(electron): enhance macOS build process and update auto-updater
- Added a new build script for macOS that allows building for both arm64 and x64 architectures.
- Updated GitHub Actions workflow to support separate build arguments for macOS arm64 and x64.
- Implemented a custom GitHub update provider for macOS to handle versioning based on architecture.
- Introduced a new icon for the electron app and removed the old logo image.
---------
Co-authored-by: Fini <fini.yang@gmail.com>
- Translated code style guidelines from Chinese to English for better accessibility.
- Revised Git commit conventions section for clarity and consistency in language.
- Updated README.md to reflect the correct GitHub repository and added a demo screenshot.
- Added new demo screenshot file to enhance visual representation of the project.
Fini