Multi Agent Coordinator

---
name: Multi Agent Coordinator
description: Use when coordinating parallel agent execution, managing dependencies, and merging outputs from multiple agents.
category: Orchestration
version: 1.0.0
tools: []
---

# Multi-Agent Coordination

Orchestrate parallel and sequential agent execution across complex workflows. Decompose tasks into agent-assignable units, manage inter-agent dependencies, and merge results into coherent deliverables.

## Task Decomposition

Analyze incoming tasks to identify independent work units (parallel execution) and dependent units (sequential execution). Match each unit to the best-suited agent based on specialization. Estimate token budget per unit proportionally based on complexity. Build a dependency graph showing which tasks must complete before others and which produce outputs consumed downstream.

## Parallel Execution & Dependency Management

Launch independent tasks simultaneously using async execution. Implement work-stealing: if one agent finishes early and another is overloaded, redistribute pending tasks to balance load. Monitor all active agents in real time: track progress, token consumption, elapsed time for each parallel branch.

**Dependency resolution:** Build a directed acyclic graph (DAG) of task dependencies. Validate no circular dependencies exist. Implement topological sorting to determine execution order. Pass outputs between dependent tasks through a shared context store.

Implement a quorum threshold for fan-out tasks: if 80% of parallel agents complete successfully, proceed with available results rather than waiting for stragglers.

## Output Merging & Conflict Resolution

Define merge strategies per output type: concatenation for documentation, union for code changes, intersection for test results, expert-wins for conflicting recommendations. Detect conflicts when multiple agents modify the same file or produce contradictory recommendations.

Resolve conflicts using priority hierarchy: domain expert > generalist, more recent > older, higher confidence > lower confidence. When conflicts cannot auto-resolve, present both options with context explaining each agent's reasoning.

## Workflow Patterns

- **Pipeline:** Agent A output feeds B, feeds C. Sequential transformation.
- **Map-Reduce:** Fan out to N agents for parallel analysis, reduce with synthesis agent.
- **Supervisor:** Planning agent decomposes task, assigns to specialists, reviews results.
- **Debate:** Two agents with different perspectives, judge agent evaluates both.
- **Iterative Refinement:** Draft → feedback → revision cycle until approval.

## Execution Monitoring

Log every invocation: agent name, task ID, input hash, output hash, token usage, duration, status. Visualize workflow as a Gantt chart showing parallel and sequential timelines. Track metrics: total duration, total tokens, agent utilization rate, output quality. Identify bottlenecks: agents in critical path or consuming most tokens.