Agent Packs

Packs are YAML-defined bundles that provision a complete supervisor/worker team with a single operation.

Pack Format

A pack is a directory containing:

packs/my-pack/
  pack.yaml              # Manifest file
  prompts/
    supervisor.txt       # Supervisor prompt
    coding_worker.txt    # Worker prompt
    ...
  extensions/
    data_platform.txt    # Domain extension prompt

pack.yaml Schema

name: My Supervisor Pack
version: "1.0"
description: >
  Multi-agent supervisor with coding and operations workers.

agents:
  - id: agent-supervisor
    name: Supervisor Agent
    prompt_file: prompts/supervisor.txt
    role: leader          # "leader" or "worker"
    order_index: 0
    engine: claude_code   # Execution engine preference

  - id: worker-coding
    name: Coding Worker
    prompt_file: prompts/coding_worker.txt
    role: worker
    order_index: 1
    engine: claude_code
    target: linux-pool
    worker_config:
      execution_engine_preference: claude_code
      worker_pool: linux_worker_pool
      mcp_servers:
        - name: filesystem
          type: stdio
          command: npx
          args: ["-y", "@anthropic-ai/mcp-filesystem"]
          description: File read/write
        - name: git
          type: stdio
          command: npx
          args: ["-y", "@anthropic-ai/mcp-git"]
      permissions:
        allow:
          - "Read(/src/**)"
          - "Bash(git diff *)"
        deny:
          - "Bash(rm -rf *)"
        ask:
          - "Bash(git push *)"
      allowed_tools: [read_file, write_file, bash, git]
      allowed_commands: [git, npm, python, pytest]

extensions:
  - id: data-platform
    name: Data Platform Extension
    prompt_file: extensions/data_platform.txt
    domain_tags:
      - "domain:data_platform"

team:
  id: supervisor-team
  name: My Supervisor Team
  mode: supervisor

Applying a Pack

Packs are applied programmatically via the PackLoader:

from supervisor.pack.loader import PackLoader

loader = PackLoader()
manifest = loader.load("packs/my-pack")
results = await loader.apply("packs/my-pack", manifest, api_client)
# results = {"prompts": [...], "agents": [...], "team": "supervisor-team"}

The apply operation:

Creates prompts for each agent and extension
Creates agent records with worker configurations
Creates the team with agent assignments

Example: Default Supervisor Pack

The agents-gateway-tests repository ships an example packs/default_supervisor/ pack that can be loaded via the CLI. It includes:

Agent	Role	Engine	Description
`agent-supervisor`	leader	claude_code	Classifies and routes requests
`worker-coding`	worker	claude_code	Repository-based coding
`worker-planning`	worker	claude_code	Read-only analysis and planning
`worker-infrastructure`	worker	claude_code	Terraform/Helm/CDK changes
`worker-operations`	worker	direct_ops	Runtime operations (kubectl, AWS)
`worker-documentation`	worker	claude_code	Documentation updates
`worker-verifier`	worker	claude_code	Test and lint verification
`worker-data-platform`	worker	claude_code	Data pipeline/DAG work