Context

End-to-end integration test that validates the full multi-agent orchestration pipeline — from task creation through delegation, decomposition, execution, and result aggregation. This test ensures all M4 components work together correctly.

Acceptance Criteria

Test Scenarios

• Scenario 1: Happy path — Lead decomposes task for 2 agents, both complete successfully, results aggregated
• Scenario 2: Partial failure — One subtask fails, parent task marked with appropriate failure status
• Scenario 3: Loop prevention — Delegation loop is detected and prevented before infinite recursion
• Scenario 4: Parallel execution — Multiple agents execute concurrently, results correctly aggregated

Test Infrastructure

• Mocked LLM (no real API calls)
• Full delegation chain validated (task flows down, results flow up)
• Status transitions validated at each step
• Loop prevention verified with clear assertion messages
• Test is deterministic and repeatable

Dependencies

• Depends on task decomposition and routing engine (Implement task decomposition and routing engine #14)
• Depends on loop prevention (Implement loop prevention (depth limits, dedup, circuit breakers, ancestry) #17)

Design Spec Reference

• Section 5 — Communication
• Section 6 — Task Workflow