pumasi

Orchestrate parallel coding with Codex CLI workers and Claude as project manager.

74|15|Updated Feb 24, 2026
One-click install
npx skills add https://github.com/fivetaku/pumasi --skill pumasi
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: pumasi
Source: https://github.com/fivetaku/pumasi/tree/main/skills/pumasi
Command: npx skills add https://github.com/fivetaku/pumasi --skill pumasi

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires yaml, and includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill tackles large-scale coding projects by offloading implementation to multiple parallel Codex instances, orchestrated by Claude as a Project Manager, significantly reducing development time and Claude's token consumption.

Core Features & Use Cases

  • Parallel Development: Distributes independent subtasks to multiple Codex CLI workers for simultaneous implementation.
  • Claude as PM: Claude designs the architecture, defines task signatures, and sets validation gates, but does not write code bodies.
  • Automated Validation: Employs bash-based "gates" (e.g., tsc, npm run build, grep) for token-free verification of subtasks.
  • Use Case: Building a new microservice with a backend API, a frontend component, and utility modules. Pumasi can assign each of these to a separate Codex worker, allowing them to be developed concurrently.

Quick Start

Use the pumasi skill to create a Todo app by describing the task.

Frequently Asked Questions about pumasi

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I use Claude as a project manager for parallel code generation?

To use Claude as a project manager for parallel code generation, this skill orchestrates Claude to design architecture and define task signatures, then delegates code body writing to multiple Codex CLI instances working simultaneously.

Can I decompose large coding tasks into independent subtasks for parallel execution?

Yes, you can decompose large coding tasks into independent subtasks for parallel execution. The skill breaks down greenfield development projects and assigns each subtask to a separate Codex worker with defined requirements.

What is the best way to reduce Claude token consumption during large-scale coding?

The best way to reduce Claude token consumption during large-scale coding is to offload implementation to parallel Codex instances. Claude designs the architecture while automated bash gates validate the code without requiring token-heavy reviews.

How do automated bash gates validate subtasks in a parallel coding workflow?

Automated bash gates validate subtasks in a parallel coding workflow by running token-free verification commands like tsc, npm run build, and grep. This ensures each Codex worker's output meets defined requirements before integration.

Does parallel coding with Codex CLI support multi-round dependencies?

Yes, parallel coding with Codex CLI supports multi-round dependencies. The orchestration handles sequential dependencies between subtasks, allowing complex greenfield development scenarios to progress through multiple gated validation rounds.

What are the limitations of using Claude as a PM for greenfield development?

A limitation of using Claude as a PM for greenfield development is that Claude only designs architecture and sets validation gates but does not write code bodies. Implementation relies entirely on Codex CLI instances executing assigned subtasks.