# agent-orchestrator

Agent Orchestrator for Solo Operators

Overview
This skill helps one operator run multi-agent execution without chaos. It breaks complex work into independent chunks, delegates clearly, and consolidates outputs into one shippable result.

The main risk in multi-agent systems is coordination overhead. This framework minimizes that risk with strict task boundaries, explicit ownership, and checkpointed consolidation.

Use with [agent-task-manager](./agent-task-manager.md), [workflow-documentation](./workflow-documentation.md), and [quality-assurance](./test-master.md).

When to Use This Skill
- A task has multiple independent streams (research, build, QA, docs).
- One person needs parallel execution in a limited time window.
- Deliverables can be validated with explicit acceptance criteria.

Core Workflow

## Step 1: Decompose the Macro Task
For each subtask, define:
- scope (what is included/excluded),
- deliverable format,
- acceptance criteria,
- dependencies.

Good decomposition reduces cross-agent coordination.

## Step 2: Assign Ownership and Interfaces
Each sub-agent must own a distinct surface:
- file paths,
- artifact names,
- output schema,
- handoff expectations.

Avoid shared write targets whenever possible.

## Step 3: Dispatch with Complete Instructions
Every agent receives:
- clear objective,
- source inputs,
- quality bar,
- deadline/checkpoint,
- escalation rule if blocked.

## Step 4: Monitor by Milestones
Monitor for outcomes, not activity:
- pending -> running -> completed/failed.
- checkpoint comment with evidence.
- blocker escalated with explicit owner.

## Step 5: Consolidate and Validate
Consolidation checklist:
- all required artifacts present,
- acceptance criteria verified,
- conflicts resolved,
- final output formatted for delivery.

Execution Template
```text
Goal: <single business outcome>
Subtask owner: <agent>
Input: <files/URLs/context>
Output: <artifact + format>
Done when: <testable criteria>
Deadline: <date/time>
Escalation: <who + when>
```

Example: Weekly Content Sprint
- Agent A: keyword/opportunity research.
- Agent B: draft updates for top pages.
- Agent C: internal links + QA pass.
- Orchestrator: final edits, publish packet, metric baseline.

Why This Works for One-Person Companies
- Parallel output without losing quality control.
- Less context switching for the operator.
- Faster cycle time from planning to publish.

Common Mistakes to Avoid
- Delegating vague tasks.
- Allowing overlapping file ownership.
- Running too many agents without clear success criteria.
- Skipping a final integration/QA gate.

Definition of Done
Orchestration is done when the final deliverable ships on time with no unresolved dependency, no ambiguous ownership, and no missing acceptance proof.