Claude Agents Root

Root configuration for Claude Code multi-agent orchestration. Defines teams, agent profiles, memory systems, skills, and commands that apply across all projects.

What is this?

This is the ~/.claude directory — the global configuration layer for Claude Code. It provides:

• Multi-agent orchestration with an orchestrator-worker pattern
• Team-based agent profiles with personality, competencies, and memory
• Reusable skills and commands shared across projects
• A 3-tier memory system (tech, project, sprint) for agent continuity

Structure

~/.claude/
├── CLAUDE.md                  # Global rules (loaded in every conversation)
├── ORCHESTRATOR.md            # Multi-agent orchestration protocol
│
├── teams/                     # Agent teams
│   ├── README.md              # Profile format & competency model
│   ├── docs/                  # Shared team documentation
│   │   ├── memory-templates.md
│   │   └── multi-agent-patterns.md
│   └── software-factory/      # Team: Software Factory
│       ├── TEAM.md            # Roster, org chart, task routing
│       ├── *.agent.md         # 20 agent profiles (VS Code standard format)
│       ├── docs/              # Team-specific protocols
│       └── agents/            # Agent memory (not profiles)
│           └── [role-name]/
│               └── memory/
│                   └── tech_memory.md
│
├── commands/                  # Slash commands (available globally)
│   └── meet.md                # Meet — Agent meeting simulation
│
└── skills/                    # Reusable skills
    ├── delegate/SKILL.md      # Delegate tasks between agents
    ├── go/SKILL.md            # Go execution skill
    └── talkwith/skill.md      # Talk to a specific agent

How it works

1. Project links to a team

Each project declares its team in [project-root]/.claude/TEAM.md:

team: software-factory

2. Orchestrator routes tasks

When you give Claude Code a task, the orchestrator:

1. Reads the project's TEAM.md to identify the team
2. Loads the team's roster and routing table
3. Assigns the task to the right agent(s)
4. Each agent reads its profile, memory, and project rules before working

Generic agents are never used when a team exists. Every task goes to a named agent with a defined profile, personality, and memory.

3. Agents have identity

Each agent has:

• MBTI personality that drives behavior under stress and in flow
• Competency matrix (technical + behavioral) at defined levels
• Voice and constraints — how they communicate and what they won't do
• 3-tier memory — tech (global), project (per-repo), sprint (temporal)

4. Memory persists across conversations

Tier	Scope	Location
Tech	Global	~/.claude/teams/[team]/agents/[name]/memory/tech_memory.md
Project	Per repo	[project-root]/.claude/memory/agents/[name]/project_memory.md
Sprint	Per sprint	[project-root]/.claude/memory/agents/[name]/sprint_N_memory.md

The Software Factory team

A full software agency with 18 agents:

Role	Agent	Model
Account Manager (entry point)	Leo	Sonnet
CEO	Alejandro	Opus
Tech Lead	Roman	Opus
Consultant	Eduardo	Opus
Talent Architect	Diana	Opus
Lead Backend	Ingrid	Sonnet
Lead Frontend	Lucas	Sonnet
Head of Design	Sofia	Sonnet
Scrum Master	Tomas	Sonnet
IT Manager	Helena	Sonnet
Ops Manager	Martin	Sonnet
Marketing Manager	Valentina	Sonnet
Module Index (Core)	Aurora	Sonnet
Module Index (Growth)	Felix	Sonnet
Dev Backend	Carmen, Raul	Haiku
Dev Frontend	Marta, Nico	Haiku

All requests enter through Leo, who triages and routes. You can bypass him by naming an agent directly.

Key principles

• No generic agents — if a team exists, every task goes to a profiled agent
• Mandatory context loading — agents must read their profile and memory before working
• Parallel by default — independent tasks run concurrently
• Memory over repetition — agents learn and don't repeat mistakes
• Spanish (Spain) for communication, English for code

License

Private configuration. Not intended for redistribution.