/parallel

command: parallel
disable-model-invocation: true
description: "[advanced] Team of Teams - Decompose compound tasks across independent claude instances"
aliases:
  - team
  - teams

Parallel - Team of Teams

When dispatching native Claude agents, use only context and effort fields exposed by the current Agent tool schema. Keep repository instructions in each brief. See `docs/CLAUDE-NATIVE-INTEGRATION.md` for compatible controls and the handoff to native dynamic workflows for large Claude-only work.

INSTRUCTIONS FOR CLAUDE

When the user invokes this command (e.g., `/octo:parallel <arguments>`):

**CORRECT - Read the explicit workflow source:**

Read ${CLAUDE_PLUGIN_ROOT:-${CODEX_PLUGIN_ROOT:-${HOME}/.claude-octopus/plugin}}/.claude/skills/flow-parallel/SKILL.md, then execute it with <user's arguments>

**INCORRECT:**

Skill(skill: "flow-parallel", ...)  ❌ Wrong! Octopus skills are model-invocation disabled
Task(subagent_type: "octo:parallel", ...)  ❌ Wrong! This is a skill, not an agent type

---

**Auto-loads the parallel skill for Team of Teams orchestration.**

Quick Usage

Describe the compound task you want decomposed:

"Build a full authentication system with OAuth, RBAC, and audit logging"
"Create a complete e-commerce platform with payments, inventory, and shipping"
"Implement CI/CD pipeline with testing, linting, and deployment stages"

What Is Parallel?

**Team of Teams** orchestration — decomposes compound tasks into independent work packages and delegates each to a separate `claude -p` process. Each process loads the full Octopus plugin, giving every work package its own Double Diamond, agents, and quality gates.

Key architectural distinction: Task tool subagents don't load plugins. Independent `claude -p` processes do.

What You Get

• Work Breakdown Structure (WBS) decomposition
• Adversarial WBS cross-check to catch missed dependencies and scope overlaps before agents launch
• Independent `claude -p` processes per work package
• Full plugin capabilities in each worker
• Parallel execution with staggered launch
• Aggregated results with exit code verification

When To Use

• Compound tasks with 3+ independent components
• Tasks that would benefit from parallel execution
• Projects where each component needs full AI capabilities
• When you want isolated, non-interfering work streams

Natural Language Examples

"Parallel build auth system with OAuth, sessions, and RBAC"
"Team up on building a dashboard with charts, filters, and export"
"Decompose and parallelize the API migration across 5 services"