Multi-Agent Parallel Execution & State Coordination

Executive Summary

This sub-scenario focuses on parallel execution, status reporting, and scheduling optimization of tasks among multiple sub-Agents after Planner output. The goal is to ensure throughput and observability for complex tasks: parallel subtask success rate ≥95%, state bus latency <1 second, blocks automatically identified and mitigated.

Scope & Guardrails

• In Scope: DAG parsing, sub-Agent distribution, context injection, state bus, scheduling optimization, result aggregation.
• Out of Scope: Failure retry/human collaboration (see recovery sub-scenario), plugin internal execution logic.
• Environment & Flags: agent-orchestrator-v2, statebus-stream, scheduler-autoscale; depends on sub-Agent registry, Kafka state bus, scheduling policy library.

Participants & Responsibilities

Scope	Repository	Layer	Responsibilities & Deliverables	Owners
orchestrator-runtime	powerx	integration	DAG Runtime, dependency management, resource inference	Agent Platform Guild
sub-agent-pool	powerx	integration	Sub-Agent registration and task claiming, context injection	Agent Platform Guild
statebus	powerx	integration	State event push, block detection, scheduling optimization	Ops Reliability Center

End-to-End Flow

1. Stage 1 – DAG Loading: Load DAG generated by Planner, compute topology order and resource requirements.
2. Stage 2 – Subtask Distribution: Push tasks to sub-Agents based on tenant, permissions, and plugin availability.
3. Stage 3 – State Sync: Sub-Agents write progress and partial results to state bus for scheduler and monitoring consumption.
4. Stage 4 – Scheduling Optimization & Aggregation: Scheduler adjusts parallelism, throttling, or reordering based on state; aggregates output after all nodes complete.

Key Interactions & Contracts

• APIs / Events: POST /internal/agent/dag/{id}/execute, EVENT agent.task.status.updated, EVENT agent.task.blocked, POST /internal/plugins/{pluginId}/invoke.
• Configs / Schemas: config/agent/subagents.yaml, config/agent/scheduler_policies.yaml, docs/standards/powerx/backend/integration/09_agent/Agent_Metrics_and_Observability.md.
• Security / Compliance: Tenant isolation, idempotent task claiming, sub-Agent credential rotation, state event masking.

Usecase Links

• UC-AGENT-EXEC-COORD-001 — Multi-Agent parallel execution and state coordination.

Acceptance Criteria

1. Subtask success rate ≥95%, state sync latency <1 second.
2. Blocked tasks detected within SLA (configurable, e.g., 30 seconds), automatically trigger reordering or scaling.
3. Aggregated results written to audit and task board, avoid duplicate execution rate >0.5%.

Telemetry & Ops

• Metrics: agent.statebus.lag_ms, agent.task.parallelism, agent.task.blocked_total, agent.result.generation_latency.
• Alerts: State latency >1s, blocked tasks >20, duplicate execution rate >0.5%.
• Observability: Grafana「Agent Execution」, Datadog agent.statebus.*, Ops task board.

Open Issues & Follow-ups

Risk/Item	Impact	Owner	ETA
Sub-Agent registry not synced with plugin new versions	Task claiming failure/rollback	Plugin Guild	2025-03-08
State event schema changes not notified downstream	Metric panel anomalies	Agent Platform Guild	2025-03-01

Appendix

• docs/scenarios/agent-orchestration/SCN-AGENT-TASK-EXEC-001.md
• docs/meta/scenarios/powerx/agent-and-automation/agent-orchestration/agent-task-execution/primary.md
• scripts/qa/dag-simulator.mjs