Magentic Orchestration (Task Ledger + Stall Detection)

What Problem It Solves

For open-domain tasks, fixed decomposition is brittle. Magentic-style orchestration:

• tracks a task ledger (implicit in messages here)
• delegates dynamically to specialists
• detects stalls (repeating the same delegation)

Core Flow

flowchart TD
  L["Task ledger / context"] --> Sel["Select next delegate"]
  Sel --> W["Specialist runs"]
  W --> Upd["Update ledger"]
  Upd --> St{"Stalled?"}
  St -->|no| Sel
  St -->|yes| R["Strategy shift / replan"] --> Sel

Evolution Path

• Generalizes: Manager-Worker (dynamic instead of fixed)
• Works best with: tracing + governance + evals (otherwise it can drift)

Repo Reference

• Code: src/agent_patterns_lab/patterns/magentic_orchestration.py
• Example: examples/65_magentic_orchestration.py
• Tests: tests/test_magentic_orchestration.py