- Test buildEditSystemPrompt basic structure (all major sections present)
- Test with workflow context including node listing and connection formatting
- Test scoped selection with restSummary (node count, type breakdown, boundary connections)
- Test empty restSummary does not show SELECTED SUBSET section
- Test createChatTools returns 3 properly structured tools
- Test execute behavior for answerQuestion, createWorkflow, and editWorkflow
- 21 test cases covering prompt building and tool creation
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Test all 5 operation types: addNode, removeNode, updateNode, addEdge, removeEdge
- Test skip scenarios for missing nodes/edges with descriptive reasons
- Test batch operations with mixed applied/skipped counts
- Test narrateOperations formatting for all operation types
- 28 test cases covering happy paths and edge cases
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Two labeled image input handles: 'A' (top 35%) and 'B' (bottom 65%)
- Real-time image collection from connected nodes (same pattern as OutputGalleryNode)
- ReactCompareSlider with horizontal-only slider (portrait=false)
- Corner labels A and B on comparison view
- Placeholder states for 0 or 1 images with connection status feedback
- nodrag nopan nowheel classes for interaction isolation
- BaseNode wrapper with customizable title and comments
- Build verified: zero TypeScript errors
- Install react-compare-slider dependency
- Add imageCompare to NodeType union
- Add ImageCompareNodeData interface (imageA, imageB fields)
- Add to WorkflowNodeData union type
- Register in nodeDefaults with 400x360 dimensions
- Add to WorkflowCanvas nodeTypes and minimap (teal color)
- Add to ConnectionDropMenu IMAGE_TARGET_OPTIONS
- Add executeWorkflow case to assign images[0] to imageA, images[1] to imageB
- Export ImageCompareNode from components/nodes/index
- Add to validation.ts DEFAULT_DIMENSIONS and createDefaultNodeData
- Template textarea with @ autocomplete dropdown
- Detects @ typing and shows matching connected variable names
- Autocomplete dropdown shows @name and prompt value preview
- Arrow keys navigate, Enter/Tab select, Escape closes
- Client-side computation of unresolvedVars from @(\w+) patterns
- Warning badge displays unresolved @tags in amber
- Live preview tooltip on hover showing resolved template
- availableVariables computed from connected prompt nodes with variableName
- Text input handle (target, left) and text output handle (source, right)
- BaseNode wrapper with title 'Prompt Constructor'
- Sync template to store on blur (same pattern as PromptNode)
- Local state prevents cursor jumping during typing
- Filtered autocomplete based on typed characters after @
- Position autocomplete relative to textarea with line height calculation
- Add variableName field to PromptNodeData (optional, backward compatible)
- Add promptConstructor to NodeType union
- Add PromptConstructorNodeData interface (template, outputText, unresolvedVars)
- Add to WorkflowNodeData union type
- Add promptConstructor to defaultNodeDimensions (340x280)
- Add createDefaultNodeData case for promptConstructor
- Modified PromptNode with @ variable naming dialog using createPortal
- Variable name validation: alphanumeric + underscore, max 30 chars
- Blue @ icon when variableName set, dimmed when not set
- Dialog shows variable name input, preview, Save/Clear buttons
- Register PromptConstructorNode in WorkflowCanvas nodeTypes
- Add promptConstructor minimap color (#f472b6 pink)
- Add to TEXT_TARGET_OPTIONS and TEXT_SOURCE_OPTIONS in ConnectionDropMenu
- Add getConnectedInputs support for promptConstructor outputText
- Add executeWorkflow case for promptConstructor:
- Finds connected prompt nodes via text edges
- Builds variable map from variableName + prompt values
- Resolves template by replacing @variableName with values
- Tracks unresolved variables in unresolvedVars array
- Stores outputText and unresolvedVars on node data
- Export PromptConstructorNode from components/nodes/index.ts
- Update duplicate defaultDimensions in WorkflowCanvas and quickstart validation
- Create OutputGalleryNode.tsx with BaseNode wrapper
- Single image input handle (type target, id image)
- Real-time image collection from connected nodes (imageInput, annotation, nanoBanana)
- 3-column scrollable thumbnail grid with nowheel for scroll isolation
- Full-screen lightbox via createPortal with:
- Left/right arrow navigation
- Download button (saves as gallery-image-N.png)
- Close button (X icon, top-right)
- Click backdrop to close
- Keyboard support (Escape, ArrowLeft, ArrowRight)
- Image counter display (N / Total)
- Empty state placeholder when no images connected
- Fix validation.ts: add outputGallery dimensions and createDefaultNodeData case
- Build passes with zero errors
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add outputGallery to NodeType union in types/nodes.ts
- Add OutputGalleryNodeData interface with images array field
- Add outputGallery to WorkflowNodeData union
- Register in nodeDefaults: dimensions 320x360, default data with empty images array
- Export OutputGalleryNode from nodes/index.ts
- Register outputGallery in WorkflowCanvas nodeTypes with pink minimap color (#ec4899)
- Add outputGallery to ConnectionDropMenu IMAGE_TARGET_OPTIONS with grid icon
- Add outputGallery case to executeWorkflow in workflowStore
- Add outputGallery to getNodeHandles (image input, no outputs)
- Add outputGallery dimensions to keyboard shortcut default dimensions
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add createdAt timestamp to WorkflowEdgeData interface
- Store createdAt on edge creation in onConnect and addEdgeWithType
- Add getImageSequenceNumber helper to EdgeToolbar
- Display "Image N" label for image connections with 2+ edges to same target
- Sequence numbers based on createdAt for stable ordering
- Single connections and text connections show no numbering
The isRunning check and set were separated by an await, allowing a second
call to pass the guard during the async gap. Move set() before the await
to close the window.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Resolve package.json conflict: take master's react-dom ^19.2.4,
keep develop's AI SDK and react-markdown additions.
Regenerate package-lock.json.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Import extractSubgraph from subgraphExtractor
- Parse selectedNodeIds from request body
- Call extractSubgraph before building workflow context
- Pass subgraph.selectedNodes and selectedEdges to buildWorkflowContext
- Update buildEditSystemPrompt to accept optional restSummary parameter
- Add WORKFLOW CONTEXT (SELECTED SUBSET) section when scoped
- Include rest-of-workflow summary with type breakdown and boundary connections
- Add metadata placeholder note to system prompt
- Enhance error handling to return 413 for oversized payloads
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Import stripBinaryData from contextBuilder
- Rewrite chatWorkflowState memo to use stripBinaryData for rich but compact node data
- Compute selectedNodeIds from nodes.filter(n => n.selected)
- Pass selectedNodeIds as new prop to ChatPanel
- Update ChatPanel to accept and forward selectedNodeIds in API requests
- Add dismissible "Focused on N selected nodes" chip above input when nodes selected
- Add error handling for oversized payloads with user-friendly message
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add 22 tests covering:
- Binary field stripping across all 8 node types
- Metadata placeholder generation with size/dimensions
- History and ref field removal
- Parameter preservation
- Position inclusion
- Enhanced buildWorkflowContext with full node data
- formatContextForPrompt rendering
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Implement extractSubgraph function with full selection logic
- Early return for no selection (return all nodes/edges, isScoped=false)
- Filter nodes into selected vs unselected with O(1) Set lookup
- Classify edges: fully within selection, boundary, or outside
- Build type breakdown from unselected nodes
- Identify boundary connections with direction (incoming/outgoing)
- All 8 tests passing (GREEN phase)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Create comprehensive test suite for extractSubgraph function
- Test cases cover: no selection, select all, single node, connected pairs,
subset selection, type breakdown, disconnected nodes, multi-handle scenarios
- All 8 tests failing with "Not implemented" error (RED phase)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add execute functions to all three chat tools (pass-through pattern)
so AI SDK produces proper tool-call/tool-result pairs in conversation
history, fixing AI_MissingToolResultsError on subsequent messages
- Fix stale closure in ChatPanel customFetch by using a ref for
workflowState, ensuring current canvas state is sent with each request
- Add EDITABLE NODE PROPERTIES section to system prompt so LLM uses
correct data property names (e.g. "prompt" not "text" for prompt nodes)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add applyEditOperations to workflowStore for AI-driven workflow modifications.
Wire ChatPanel to process editWorkflow tool calls and apply operations via store.
Pass lightweight workflow state to chat API for context-aware assistance.
Changes:
- workflowStore: add applyEditOperations method and EditOperation import
- ChatPanel: add onApplyEdits prop, workflowState prop, custom fetch wrapper
- ChatPanel: process tool-invocation parts for editWorkflow and createWorkflow
- ChatPanel: render tool explanations in chat UI
- WorkflowCanvas: extract chatWorkflowState, create handleApplyEdits callback
- WorkflowCanvas: pass onApplyEdits and workflowState to ChatPanel
AI edits trigger captureSnapshot before applying (revert button support).
AI edits do NOT increment manualChangeCount (only manual edits clear snapshot).
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Create SUMMARY.md documenting plan execution and update STATE.md
with current position and accumulated decisions.
Summary includes:
- Task completion: enhanced chat API route with tool calling
- Deviation: changed maxSteps to stopWhen: stepCountIs(3) (AI SDK v6 API)
- Key decisions: tool pattern, context injection, multi-step reasoning
- Verification: TypeScript passes, streaming works correctly
STATE.md updates:
- Current position: Plan 2 of phase 34
- Performance metrics: 2/? plans, 13 min total, 6.5 min avg
- Recent trend: 5-8 min execution
- Accumulated context: 4 new decisions about chat API and tool calling
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Broaden the trailing-slash exemption to recognize both C:\ and C:/ as
drive roots. Tighten the client-side absolute path check to require a
separator after the drive letter and accept UNC paths (\\server\share).
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Rewrite /api/chat route to accept workflow context from client and use
AI SDK tool calling for multi-modal agent routing.
Changes:
- Accept optional workflowState (nodes/edges) in request body
- Build workflow context using buildWorkflowContext from contextBuilder
- Generate context-aware system prompt using buildEditSystemPrompt
- Create chat tools (answerQuestion, createWorkflow, editWorkflow) with
current node IDs for validation
- Configure streamText with tools, toolChoice: 'auto', and stopWhen for
multi-step reasoning (up to 3 steps)
- Stream results back via toUIMessageStreamResponse for useChat hook
compatibility
LLM now automatically routes to the correct tool based on user intent.
Tool results stream back to client for application in Plan 03.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Create tools.ts with buildEditSystemPrompt function that constructs enhanced system prompt with workflow context and tool usage rules. Implement createChatTools function that returns three tool definitions using AI SDK v6 tool() pattern:
- answerQuestion: informational queries without workflow modification
- createWorkflow: build new workflows from scratch via quickstart
- editWorkflow: targeted edits with operation list and explanation
All tools use zod schemas for type-safe input validation. Tools follow "generate" pattern (no execute function) - LLM provides structured output that client/route handles.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Create editOperations.ts with EditOperation type union covering all 5 operation types (addNode, removeNode, updateNode, addEdge, removeEdge). Implement applyEditOperations function with batched immutable updates, proper ID generation, and validation with skip tracking. Add narrateOperations function for human-readable summaries.
Create contextBuilder.ts with WorkflowContext interface and buildWorkflowContext function that strips base64 data, history arrays, and internal state. Add formatContextForPrompt function for system prompt injection.
Both modules are pure TypeScript with no UI or API dependencies.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
On macOS, osascript's folder picker can return hostname-prefixed paths
(e.g. "HOSTNAME/Users/...") for network-mounted volumes instead of
absolute POSIX paths, causing directory validation to fail.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Use neutral-300 text, neutral-700/50 bg, and neutral-600 border
to make the button noticeable without clashing with the app style.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
loadWorkflow was unconditionally calling clearSnapshot(), wiping the
snapshot that captureSnapshot() just saved. Add preserveSnapshot
option to loadWorkflow and pass it from the AI build workflow path
so the Revert AI Changes button actually appears.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Updated last activity date and details in STATE.md to reflect completion of quick-001.
- Added recent decisions regarding dynamicInputs type and behavior.
- Enhanced tests in QuickstartTemplatesView to handle new download URL structure for community workflows, ensuring proper response handling for both API and direct download scenarios.
Update vulnerable react-dom dependency and next.js to patched versions.
Fix QuickstartTemplatesView test to mock two-step R2 download flow.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Test: multiple images to same schema-mapped handle aggregate into array
- Test: single image to schema-mapped handle stays as plain string
- Test: multiple images with distinct schema handles remain separate strings
- Test: three images to same handle produce array of length 3
- Fix react-dom version mismatch (19.2.0 -> 19.2.3) to match react
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Update dynamicInputs type from Record<string, string> to Record<string, string | string[]> in 4 locations
- Fix getConnectedInputs to aggregate same-key values into arrays instead of overwriting
- Fix dynamicInputs.prompt access to handle array case
- Fix hasPrompt check in POST handler to handle array-typed prompt
- Update processedDynamicInputs type in both Replicate and fal blocks
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Test: multiple images to same schema-mapped handle aggregate into array
- Test: single image to schema-mapped handle stays as plain string
- Test: multiple images with distinct schema handles remain separate strings
- Test: three images to same handle produce array of length 3
- Fix react-dom version mismatch (19.2.0 -> 19.2.3) to match react
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Update dynamicInputs type from Record<string, string> to Record<string, string | string[]> in 4 locations
- Fix getConnectedInputs to aggregate same-key values into arrays instead of overwriting
- Fix dynamicInputs.prompt access to handle array case
- Fix hasPrompt check in POST handler to handle array-typed prompt
- Update processedDynamicInputs type in both Replicate and fal blocks
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The proxy was downloading 80-275MB workflow files through a serverless
function, causing timeouts and failures. Now the proxy just returns the
presigned R2 URL and the client downloads directly from R2.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Kie API returns HTTP 200 even on errors, with error code in the JSON body.
Now properly checks createResult.code !== 200 and returns the actual error
message from the API instead of generic "No task ID in response".
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Some Kie.ai models (like gpt-image/1.5-text-to-image) require certain
parameters to be present even if not user-specified (e.g., aspect_ratio,
quality). Add getKieModelDefaults() function that provides model-specific
defaults which are merged with user-provided parameters.
Also skip file logging on Vercel since filesystem is read-only.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Use multipart/form-data for file upload with 'file' and 'uploadPath' fields
- Extract download URL from data.downloadUrl in upload response
- Wrap parameters in 'input' object for standard models
- Use flat structure with camelCase for Veo models
- Extract results from resultJson.resultUrls in poll response
Co-Authored-By: Claude <noreply@anthropic.com>
- Add Kie.ai section in provider settings with API key input
- Add kie to env-status API for .env detection
Co-Authored-By: Claude <noreply@anthropic.com>
shrimbly