Kudosflow2 v1.3.0 - Agent Teams Fan-out/Fan-in Parallel Execution

What's New in v1.3.0

Added

• Agent Teams Fan-out/Fan-in: Replaced sequential A2A worker execution with LangGraph-native parallel execution
  • New planner_node: Analyzes user task and outputs a JSON array of worker definitions (name, role, task)
  • New worker_node: Executes each worker as an independent LangGraph node in the same process
  • New aggregator_node: Collects all worker results and generates an integrated report
  • New finalize_node: Presents final report with confirmation prompt before cleanup
  • Workers now run in parallel via LangGraph Send API (fan-out), results merged via fan-in

Changed

• Teams leader.json: Fully redesigned graph structure from single leader+tools nodes to planner→worker×N→aggregator→finalize pipeline (version 1.0.0 → 2.2.0)
• Teams leader.json state: Replaced teamPlan: string with workerPlans: {name, role, task}[], currentPlan, workerResults, finalReport for fan-out/fan-in state management
• WorkflowExecutor: Output resolution now prefers finalReport state field when present, falling back to last assistant message — supports fan-out/fan-in workflows that store results outside the message list
• extractPossibleTargets: Added new Send('nodeName', ...) pattern detection alongside existing return 'string' pattern to support LangGraph fan-out edge declarations

Removed

• Teams leader.json: Removed tools_node (ToolNode) — worker tasks are now executed as LangGraph nodes, not via bash/A2A tool calls
• Teams leader systemPrompt: Removed embedded 8-step A2A orchestration instructions (worker JSON write → launch → healthcheck → send → read results)

Installation

```bash
code --install-extension kudosflow2-1.3.0.vsix
```