skills.rest
buildoak
buildoakCOMMUNITY
Explore all skills by this author →
GitHub

codex-subagent

Spawn a Codex-based subagent for code verification, debugging, and refactoring tasks.

1|Updated Feb 8, 2026
Search Tags:#typescript#codex#multi-model#parallel#subagent#code-review
One-click install
npx skills add https://github.com/buildoak/codex-task-subagent --skill codex-subagent-buildoak
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: codex-subagent
Source: https://github.com/buildoak/codex-task-subagent/tree/main/skill/codex-subagent
Command: npx skills add https://github.com/buildoak/codex-task-subagent --skill codex-subagent-buildoak

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires @openai/codex-sdk.

What problem does it solve?

Codex-subagent provides a self-contained Codex-based worker that can be spawned by Claude to perform code-focused tasks, enabling cross-model verification, parallel execution, and nested subagent workflows.

Core Features & Use Cases

  • Verification & review: Different models catch different bugs and issues, enabling cross-model validation for code quality.
  • Implementation & generation: Codex-driven code changes, refactoring, and large-scale transformations with a fresh perspective.
  • Debugging & analysis: When Claude stalls, Codex can examine code areas independently and surface edge cases.
  • Refactoring & codebase exploration: Systematic codebase exploration and large-scale transforms across modules.
  • Nested subagent work: Claude subagents can spawn Codex workers to augment analysis or execution.
  • The 10x pattern: Codex generates/refactors at scale while Claude reviews for hidden issues and improvements.

Quick Start

Prompt Codex with your task using bun run src/codex-agent.ts and provide a clear prompt describing the code task.

Frequently Asked Questions about codex-subagent

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

FAQPage Schema
How do I use a multi-model subagent for parallel code review and verification?▼

Codex subagent enables multi-model code review by spawning a focused worker that independently examines code areas, catching bugs and issues that a single model might miss. This cross-model verification improves overall code quality.

What is the best way to debug code when Claude stalls or misses edge cases?▼

Spawning a Codex worker to independently examine the codebase and surface hidden edge cases provides a fresh second-model perspective. This approach breaks through analytical blocks when the primary model stalls.

How do I run large-scale refactoring and codebase exploration with multiple AI models?▼

Large-scale refactoring uses a Codex-driven worker to perform systematic codebase exploration and transformations across modules. The 10x pattern applies here, where Codex generates code at scale while Claude reviews for hidden issues.

Does the codex-subagent require the OpenAI Codex SDK to work?▼

The OpenAI Codex SDK is required as a core dependency to spawn and communicate with the subagent. You need the SDK installed to execute code-focused tasks via the bun run command.

Can I control the sandbox mode and model selection when spawning a parallel subagent?▼

CLI flags allow you to control sandbox mode, model selection, reasoning levels, and MCP clusters when spawning the subagent. This fine-tunes the execution environment for specific code tasks.

When should I not use a nested Claude-Codex workflow for code generation?▼

Nested Claude-Codex workflows add unnecessary overhead for simple, single-model tasks. They are best suited for complex implementation, debugging, and refactoring requiring a fresh second-model perspective.

Related Skills

Explore More Software Engineering Skills→
aimskraimskr

codex

Automate Codex CLI exec commands with model, reasoning, and sandbox flags.

Official
Advanced
gtappsgtapps
69

test-run

Run hermit test suites and report pass/fail summaries.

Community
Intermediate
CodingHeaderCodingHeader

dispatching-parallel-agents

Coordinate LLM agents to work in parallel on independent software-engineering tasks, aggregating their results and monitoring progress along the way.

Community
Intermediate
vinhnxvvinhnxv
6

codex-review

Compare Claude Code and Codex findings across code changes.

Community
Advanced