When a node has multiple named outputs, the connection lines (edges)
leaving each output handle now display the output name as a label.
Labels are applied in all edge creation paths: manual connections,
flow import, and multi-node copy-paste. Edge label CSS counters the
existing color inversion filter so labels render with normal colors.

https://claude.ai/code/session_01M94qqeYWUGVGG7cpJ7z7G3

Include node_reference in the backend NodeInput schema and
get_node_input() serialization so the frontend receives it.
Use node_reference as the edge label when the source node has
one set (e.g., "orders", "clients" connecting into a python_script).
Named output labels (e.g., "valid", "invalid") take precedence
over node_reference for multi-output nodes.

https://claude.ai/code/session_01M94qqeYWUGVGG7cpJ7z7G3

The resolve_node_paths() method hardcoded all input parquet files under
the "main" key, so flowfile.read_input("orders") failed in the
interactive notebook editor. The files are named {name}_{index}.parquet
(e.g. orders_0.parquet), so parse the input name from each filename
and group paths accordingly. A "main" key is still added as a
backward-compatible alias pointing to all input files.

https://claude.ai/code/session_01M94qqeYWUGVGG7cpJ7z7G3

- Edge labels now update reactively when node_reference changes
  (no UI refresh needed)
- Add "Show edge labels" toggle in Execution Settings dialog,
  persisted in FlowSettings (show_edge_labels field)
- When no node_reference is set, edges show df_{nodeId} as the
  default label instead of being blank
- Toggling edge labels off/on immediately hides/shows all labels

https://claude.ai/code/session_01M94qqeYWUGVGG7cpJ7z7G3

- Add label? and nodeReference? to NodeData interface in NodeWrapper
  so handle labels are recognized by the type checker
- Cast sourceNode.data to Record<string, any> in computeEdgeLabel
  since VueFlow's generic node data type defaults to {}

https://claude.ai/code/session_01M94qqeYWUGVGG7cpJ7z7G3

…/input_names endpoint

Improves coverage on kernel/execution.py (named/unnamed input paths, error handling),
input_schema.py (output_names defaults and roundtrip), and routes.py (/node/input_names
with reference, fallback, empty inputs, and 404 cases).

https://claude.ai/code/session_01M94qqeYWUGVGG7cpJ7z7G3

Backend:
- Add null guard on node before attribute access in add_python_script
- Extract _resolve_input_names helper to eliminate duplicated logic
- Use zip(strict=True) to catch input name/table count mismatches
- Use model_copy(update=) instead of dump/re-instantiate for template override
- Create NodeInputNameInfo Pydantic model for /node/input_names endpoint
- Update FlowSettings docstring for show_edge_labels field
- Move lazy imports to top-level in test file

Frontend:
- Create shared NodeHandle interface in flow.types.ts
- Create shared InputNameInfo interface in flow.types.ts
- Replace all (o: any) callbacks with typed NodeHandle finds
- Replace Record<string, any> cast with typed FlowNodeData interface
- Type onConnect params with Connection from @vue-flow/core
- Use proper types in NodeWrapper NodeData interface
- Add parseInt radix parameter
- Use unknown instead of any in catch clause

https://claude.ai/code/session_01M94qqeYWUGVGG7cpJ7z7G3

…ibility

Connection.sourceHandle/targetHandle are string|null but NodeConnection
expects specific literal types. Cast them explicitly and use ?? undefined
for the computeEdgeLabel call.

https://claude.ai/code/session_01M94qqeYWUGVGG7cpJ7z7G3

…dwardvaneechoud/Flowfile into claude/named-node-inputs-outputs-SIjmK

…w_edge_labels default mismatch. Add safe identifier validation for node_reference and output_names, defense-in-depth checks in write_inputs_to_parquet, and graceful fallback when input counts diverge.