Do I need any additional setup for this skill?

Yes, this skill requires the codex-builder MCP server to be configured and the LifeOS workflow scripts (like start_build.py) to be present in your repository.

What happens after Codex finishes its task?

Claude automatically initiates a tiered review protocol. It checks for obvious issues, applies patterned fixes, runs tests, and provides a structured report of the results and remaining items.

Can I use Codex for read-only tasks like code analysis?

Absolutely. For read-only tasks such as reviews or planning, you can set the sandbox parameter to 'read-only' which bypasses the requirement for worktree isolation.

How does this skill ensure repository safety?

The skill enforces a 'Hard Gate' that requires Codex to run in an isolated Git worktree. This prevents merge conflicts, stash failures, and accidental overwrites in your primary codebase.

What is the primary benefit of using the Codex skill?

It allows Claude to delegate specific, bounded tasks to a secondary Codex agent, providing an independent implementation attempt while maintaining repository safety through worktree isolation.

Codex Agent Delegation

Name: Codex Agent Delegation
Author: marcusglee11

bymarcusglee11

0•

生産性とワークフロー

Delegates bounded coding tasks to the Codex agent via MCP with automated context enrichment and safety-gated worktree isolation.

Codex Agent Delegation enables Claude to offload specific implementation tasks to a secondary Codex agent through an MCP integration. The skill manages the end-to-end delegation lifecycle: it gathers deep repository context from Git logs and system state files, enriches the task prompt for maximum accuracy, and enforces strict worktree isolation to prevent merge conflicts in your primary repository. Once Codex completes the task, the skill triggers a tiered review-build protocol to validate the output, apply necessary fixes, and ensure the code meets project standards before merging.

主な機能

01Full pipeline support including handoff, review-fix, and close-build stages

02Seamless MCP integration with the codex-builder server

03Hard-gated worktree isolation to prevent primary repository corruption

04Automated context enrichment using Git history and LIFEOS_STATE files

05Tiered review-build protocol for validating agent-generated code

060 GitHub stars

ユースケース

01Offloading specific feature implementations to a secondary AI model for independent execution

02Creating isolated development environments for experimental code changes

03Automating the review and fix cycle for agent-generated pull requests

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add marcusglee11/lifeos codex

For use in Claude.ai and ChatGPT