agent-orchestrator_skill

Overview

This skill implements a meta-agent orchestrator to orchestrate complex work by decomposing macro tasks, spawning autonomous sub-agents, and consolidating results. It acts as an agent factory that delegates tasks to specialized sub-agents and manages agent coordination, lifecycle, and cleanup. Designed for multi-agent workflows, it supports parallel agents and sequential dependency graphs to deliver predictable outcomes.

How this skill works

The orchestrator analyzes a macro task and applies a task breakdown to decompose task into independent subtasks with clear success criteria. For each subtask it generates a sub-agent workspace, writes instructions and inputs, and spawns agents to run autonomously; agents communicate via file-based inbox/outbox and update a status file. The orchestrator monitors status.json checkpoints, collects and validates outputs, resolves conflicts, consolidates deliverables, and dissolves agents when done.

When to use it

• When you need a meta-agent to orchestrate multi-agent workflows and delegate tasks across specialized sub-agents
• When a large task can be decomposed into parallel agents to speed delivery
• When clear agent coordination and file-based handoffs simplify integration between autonomous workers
• When you need to run isolated, auditable agent workspaces for archival or backup
• When you want to prototype an agent factory that spawns, monitors, and dissolves sub-agents automatically

Best practices

• Decompose task into subtasks that are completable in isolation and include explicit success criteria
• Prefer broader autonomous sub-agents over many tightly coupled ones to reduce inter-agent dependencies
• Use structured file handoffs (inbox/outbox and status files) for explicit agent coordination and retries
• Start small (2–3 agents) and iterate on templates before scaling to many parallel agents
• Log status and artifacts for each agent to enable easy validation, conflict resolution, and audits

Example use cases

• Research report pipeline: decompose into data-collector, analyst, writer, and reviewer agents with sequential dependencies
• Large codebase migration: spawn code agents to refactor modules in parallel and an integration agent to merge changes
• Market analysis: spawn parallel data-gathering agents for regions, then an analysis agent to consolidate findings
• QA orchestration: spawn multiple test agents against different environments, then consolidate test reports
• Content production: delegate topic research, drafting, and editing to specialized writer and reviewer sub-agents