team-load

Team Load — Bootloader Skill

Purpose

Load a team pack folder and bootstrap it as a running Claude Code agent team. Parse the universal team.yaml format, resolve agent definitions, inject shared rules, map tools, and spawn all agents.

When to Use

This skill activates when:

• User asks to load, start, or bootstrap a team from a folder
• User references a team pack by name or path
• User wants to create a team from a saved configuration
• The /team-load command is invoked

Team Pack Format

A team pack is a folder with this structure:

team-name/
├── team.yaml              # Required: team manifest
├── orchestrator.md        # Required: lead agent definition
├── agents/
│   ├── agent-1.md         # Agent definitions
│   └── agent-2.md
├── workflows/
│   └── workflow-1.yaml    # Task sequences
└── rules/
    └── standards.md       # Shared rules

Bootload Process

Step 1: Parse team.yaml

Read the manifest file. Required fields:

name: team-name
version: "1.0"
description: What the team does

orchestrator:
  file: orchestrator.md
  model: opus

agents:
  - name: agent-name
    file: agents/agent-name.md
    model: sonnet
    tools: [file_read, file_write, shell]
    reports_to: orchestrator

shared_rules:
  - rules/standards.md

tool_mappings:
  claude_code:
    file_read: Read
    file_write: Write
    file_edit: Edit
    shell: Bash
    web_search: WebSearch
    file_search: Glob
    content_search: Grep

Step 2: Load Shared Rules

Read each file listed in shared_rules and concatenate into a single rules block.

Step 3: Load Orchestrator

1. Read the orchestrator's .md file
2. Strip YAML frontmatter (everything between --- markers)
3. Build the team roster string from agents list
4. Replace {{TEAM_ROSTER}} in the prompt with the roster
5. Append shared rules under ## Shared Rules heading

Step 4: Load Agent Definitions

For each agent in agents list:

1. Read the agent's .md file from the path in file
2. Strip YAML frontmatter, keep body as system prompt
3. Append shared rules under ## Shared Rules heading
4. Map tools using tool_mappings.claude_code

Step 5: Create Claude Code Team

TeamCreate({
  team_name: config.name,
  description: config.description
})

Step 6: Spawn Agents

For each agent, spawn as a teammate:

Task({
  team_name: config.name,
  name: agent.name,
  subagent_type: "general-purpose",
  model: agent.model,
  prompt: processedPrompt,
  run_in_background: true
})

Spawn ALL agents in a single message for parallel startup.

Step 7: Report

Output a summary table:

Agent	Model	Role	Status
name	model	role	spawned

List available workflows and how to run them.

Tool Mapping

The tool_mappings section in team.yaml translates generic tool names to runtime-specific ones:

Generic	Claude Code	OpenAI	Generic
file_read	Read	file_search	read_file
file_write	Write	—	write_file
file_edit	Edit	—	edit_file
shell	Bash	code_interpreter	execute_command
web_search	WebSearch	web_search	web_search
file_search	Glob	—	find_files
content_search	Grep	—	search_content

Search Locations

When loading by name (not path), check in order:

1. ~/.claude/team-packs/{name}/
2. ./team-packs/{name}/
3. ${CLAUDE_PLUGIN_ROOT}/assets/team-packs/{name}/

Error Handling

• Missing team.yaml → error with message about required format
• Missing agent file → warn and skip that agent
• Missing shared rules file → warn and continue without rules
• Invalid YAML → error with parse details
• No tool_mappings → use generic tool names as-is

Additional Resources

For the complete team.yaml specification, see references/team-yaml-spec.md. For bootloader implementation examples, see scripts/.