# Agent Definitions

This directory contains agent persona definitions for the multi-agent system.

## Purpose

Agent files define specialized personas that Claude can adopt when working on specific domains. They are:

- **Persona documents** - Define behavior, responsibilities, and constraints
- **Reference material** - Loaded into context when the domain is relevant
- **Subagent templates** - Can be spawned via Task tool with explicit prompting

## Active Agents

| Agent | Model | Domain |
|-------|-------|--------|
| `personal-assistant` | Opus | User interface, top-level oversight |
| `master-orchestrator` | Opus | Cross-agent coordination, policy enforcement |
| `linux-sysadmin` | Sonnet | Workstation management |
| `k8s-orchestrator` | Opus | Kubernetes cluster management |
| `programmer-orchestrator` | Opus | Code development coordination |

## Supporting Agents (K8s)

| Agent | Model | Domain |
|-------|-------|--------|
| `k8s-diagnostician` | Sonnet | Cluster health, pod diagnostics |
| `argocd-operator` | Sonnet | GitOps deployments |
| `prometheus-analyst` | Sonnet | Metrics and alerting |
| `git-operator` | Sonnet | Repository operations |

## Supporting Agents (Programming)

| Agent | Model | Domain |
|-------|-------|--------|
| `code-planner` | Sonnet | Design and architecture |
| `code-implementer` | Sonnet | Code writing |
| `code-reviewer` | Sonnet | Code review and quality |

## Agent File Format

```yaml
---
name: agent-name
description: When to use this agent
model: sonnet|opus|haiku
tools: Tool1, Tool2, ...
---

# Agent Name

[Instructions in Markdown]
```

## Note on Execution

Claude Code doesn't automatically spawn agents from this hierarchy. To use an agent:

1. **Skill-based activation** - If a skill description matches, Claude uses that context
2. **Explicit Task tool** - Use Task tool with agent prompt to spawn subagent
3. **Manual prompt** - Reference agent file directly

The hierarchy is primarily for organization and documentation.