skills.rest
mcncarl
mcncarlCOMMUNITY
Explore all skills by this author →
GitHub

yichen-codex-chatgpt

Orchestrates a ChatGPT Pro and Codex plan-implement-review loop over a read-only MCP tunnel.

2.0k|274|Updated Feb 11, 2026
Search Tags:#mcp#chatgpt#CODEX# orchestration#Code review#Multi-Agent#secure tunnel
One-click install
npx skills add https://github.com/mcncarl/yichen-skills --skill yichen-codex-chatgpt
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: yichen-codex-chatgpt
Source: https://github.com/mcncarl/yichen-skills/tree/main/yichen-codex-chatgpt
Command: npx skills add https://github.com/mcncarl/yichen-skills --skill yichen-codex-chatgpt

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Coordinating two AI agents on one coding task often degrades into manual copy-paste of code, diffs, and logs between chat windows, with no proof of which model actually reviewed what. This Skill enforces a strict protocol where ChatGPT Pro researches, plans, and reviews through a read-only MCP tunnel while Codex alone edits files and runs tests locally.

Core Features & Use Cases

  • Five routing modes: research, code, hybrid, review, and an opt-in local-only mode, selected automatically from the request.
  • C2C state machine: a fixed PRECHECK → INIT → PLAN → EXECUTING → EXECUTED → REVIEW protocol with unique task IDs, message markers, and a two-iteration execution budget.
  • Read-only evidence boundary: ChatGPT accesses one validated project through exactly seven read-only MCP tools; code, diffs, and secrets never enter the browser composer.
  • Use Case: Ask Codex to refactor a module; the Skill has ChatGPT Pro inspect the real workspace through the tunnel, return an architecture plan, let Codex implement and test it, then have ChatGPT review the actual diff before declaring DONE.

Quick Start

Ask Codex to use $yichen-codex-chatgpt to have ChatGPT Pro design and review the changes for your current project.

Frequently Asked Questions about yichen-codex-chatgpt

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

FAQPage Schema
How do I set up a ChatGPT and Codex collaboration loop for code review?▼

Install the Skill, copy config.example.md to a private location, and configure a compatible read-only MCP runtime per references/setup.md. Codex then sends protocol messages to ChatGPT Pro, which inspects your project through the tunnel and returns PLAN and REVIEW responses.

What MCP tools does ChatGPT use to inspect my project?▼

ChatGPT accesses exactly seven read-only tools: workspace_overview, list_files, search_code, read_snippet, review_changes, verification_report, and review_packet. The tunnel never exposes write, delete, command execution, or web search tools to ChatGPT.

Does this Skill include the MCP runtime or tunnel client?▼

No. The public package contains only the orchestration protocol. You must supply a compatible runtime satisfying the contract in references/setup.md; the author's tunnel client, Runtime Key, and private ChatGPT App are not published.

Can ChatGPT edit files or run tests in this workflow?▼

No. ChatGPT only researches, plans, and reviews through read-only tools. Codex is the sole writer and command executor, and the protocol forbids claiming ChatGPT modified files or ran tests.

What happens if ChatGPT keeps requesting changes after review?▼

The protocol allows at most two execution iterations per task ID. If the second review still requests changes, the task ends as BLOCKED and further repair requires explicit user authorization under a new task ID.

Why does the Skill require ChatGPT Pro mode instead of Work mode?▼

The protocol verifies a visible Chat/聊天 selector with an active Pro model route before every send. Work mode, API access, or other accounts invalidate the run because they cannot satisfy the evidence and private-App gates.

Related Skills

Explore More Software Engineering Skills→
leto-labsleto-labs

webdav

Expose agent workspace files via a WebDAV bridge using a Cloudflare Tunnel.

Official
Intermediate
Strategic-AutomationStrategic-Automation
6

requesting-code-review

Automate pre-commit verification with security scans, quality gates, and reviewer subagent.

Official
Advanced
coinvest518coinvest518

web-research

Automate multi-source web research and generate cited reports via delegated subagents.

Community
Intermediate
gevreygevrey

chatgpt-browser-access

Verify ChatGPT login status and fetch recent conversations in the OpenClaw browser.

Community
Advanced