Skip to content

Self-improving AI Employees for everyone

feat(client): three-tier schema dispatch in InputSection (run data > backend > legacy)

trohitg327f792f48f9
diff --git a/client/src/components/parameterPanel/InputSection.tsx b/client/src/components/parameterPanel/InputSection.tsx
index f06fac01..6e1e1d8b 100644
--- a/client/src/components/parameterPanel/InputSection.tsx
+++ b/client/src/components/parameterPanel/InputSection.tsx
@@ -5,6 +5,75 @@ import { useWebSocket } from '../../contexts/WebSocketContext';
import { Node, Edge } from 'reactflow';
import { useDragVariable } from '../../hooks/useDragVariable';
import { useAppTheme } from '../../hooks/useAppTheme';
+import { queryClient } from '../../lib/queryClient';
+
+// ---------------------------------------------------------------------------
+// Backend-driven node output schema lookup.
+//
+// Mirrors n8n's schemaPreview pattern (see
+// docs-internal/schema_source_of_truth_rfc.md): the shape shown in the
+// drag-drop variable panel for a node that has not been executed yet
+// comes from the backend's Pydantic model registry via the
+// `get_node_output_schema` WS handler. Results are cached per node
+// type in the shared TanStack Query client (in-memory only, matching
+// n8n's approach — schemas are small and the cache is cheap).
+//
+// Kept as a plain async helper (not a hook) because InputSection
+// iterates over connected edges synchronously during render build-up
+// and then needs to populate schemas for each unique node type once.
+// ---------------------------------------------------------------------------
+
+type NodeOutputSchema = Record<string, any> | null;
+
+const nodeOutputSchemaQueryKey = (nodeType: string) =>
+ ['nodeOutputSchema', nodeType] as const;
+
+async function fetchNodeOutputSchema(
+ nodeType: string,
+ sendRequest: (type: string, data: any) => Promise<any>,
+): Promise<NodeOutputSchema> {
+ return queryClient.fetchQuery({
+ queryKey: nodeOutputSchemaQueryKey(nodeType),
+ queryFn: async () => {
+ try {
+ const response = await sendRequest('get_node_output_schema', { node_type: nodeType });
+ return (response?.schema ?? null) as NodeOutputSchema;
+ } catch {
+ return null;
+ }
+ },
+ staleTime: Infinity,
+ });
+}
+
+/**
+ * Flatten a JSON-Schema-7 "object" shape into the plain
+ * { field: 'primitive-type-name' | nestedObject } map the variable
+ * panel expects. Only the fields the UI actually needs; other JSON
+ * Schema keywords (minLength, pattern, etc.) are ignored.
+ */
+function jsonSchemaToShape(schema: Record<string, any> | null | undefined): Record<string, any> | null {
+ if (!schema || typeof schema !== 'object') return null;
+ const props = schema.properties;
+ if (!props || typeof props !== 'object') return null;
+
+ const out: Record<string, any> = {};
+ for (const [key, raw] of Object.entries(props)) {
+ if (!raw || typeof raw !== 'object') continue;
+ const prop = raw as Record<string, any>;
+ if (prop.type === 'object' && prop.properties) {
+ out[key] = jsonSchemaToShape(prop) ?? 'object';
+ } else if (prop.type === 'array') {
+ out[key] = 'array';
+ } else if (typeof prop.type === 'string') {
+ // string | number | integer | boolean | null
+ out[key] = prop.type === 'integer' ? 'number' : prop.type;
+ } else {
+ out[key] = 'any';
+ }
+ }
+ return out;
+}
interface InputSectionProps {
nodeId: string;
@@ -39,7 +108,7 @@ const renderNodeIcon = (icon: string, size: number = 16) => {
const InputSection: React.FC<InputSectionProps> = ({ nodeId, visible = true }) => {
const theme = useAppTheme();
const { currentWorkflow } = useAppStore();
- const { getNodeOutput } = useWebSocket();
+ const { getNodeOutput, sendRequest } = useWebSocket();
const [expandedNodes, setExpandedNodes] = useState<Set<string>>(new Set());
const [connectedNodes, setConnectedNodes] = useState<NodeData[]>([]);
const [loading, setLoading] = useState(false);
@@ -175,11 +244,20 @@ const InputSection: React.FC<InputSectionProps> = ({ nodeId, visible = true }) =
} else {
hasExecutionData = false;
- // No real run data yet. The start node derives its shape from the
- // user's initialData JSON (genuinely UI-owned). Every other node
- // falls through to the legacy sampleSchemas map below as a bridge
- // until the backend schema endpoint ships. See
- // docs-internal/schema_source_of_truth_rfc.md.
+ // Schema precedence (mirrors n8n VirtualSchema.vue — see
+ // docs-internal/schema_source_of_truth_rfc.md):
+ // 1. real run data (handled above in the `if` branch)
+ // 2. backend-declared schema via get_node_output_schema
+ // 3. legacy sampleSchemas map (removed when coverage complete)
+ // 4. `{ data: 'any' }` empty fallback
+ //
+ // `start` is special-cased: its shape is authored inline by the
+ // user as initialData JSON, so there's no backend schema to
+ // defer to.
+ const backendSchema = nodeType !== 'start'
+ ? jsonSchemaToShape(await fetchNodeOutputSchema(nodeType, sendRequest))
+ : null;
+
if (nodeType === 'start') {
try {
const initialData = sourceNode?.data?.initialData || '{}';
@@ -187,6 +265,8 @@ const InputSection: React.FC<InputSectionProps> = ({ nodeId, visible = true }) =
} catch (e) {
outputSchema = {};
}
+ } else if (backendSchema) {
+ outputSchema = backendSchema;
} else {
const sampleSchemas: Record<string, Record<string, any>> = {
location: { latitude: 'number', longitude: 'number', accuracy: 'number', provider: 'string', altitude: 'number' },