Code review with Codex
Codex, OpenAI's coding agent, in CodePanel's review loop: it connects through MCP and skills, reads AGENTS.md and answers every finding.
Codex is OpenAI's coding agent, in a CLI, an extension for VS Code and compatible editors, the ChatGPT desktop app and Codex cloud, started from ChatGPT. With CodePanel it is the author on the other side of the review: after it pushes, it reads the panel's findings on that commit over MCP, fixes or disputes each one with evidence, and rates every finding. It reads AGENTS.md and AGENTS.override.md, and the setup skill writes the loop into AGENTS.md or CLAUDE.md.
How Codex connects
Codex connects through an MCP server entry in ~/.codex/config.toml and the codepanel and codepanel-setup skills. The steps are in Codex.
Codex features for the review loop
| Feature | What the docs say | Source |
|---|---|---|
| Maker | OpenAI | github.com |
| Runs in | a CLI, an extension for VS Code and compatible editors, the ChatGPT desktop app and Codex cloud, started from ChatGPT | learn.chatgpt.com, page 2, page 3, page 4 |
| MCP | Remote servers over streamable HTTP with OAuth (codex mcp login), configured in config.toml |
learn.chatgpt.com |
| Hooks | Scripts or MCP tools in hooks.json or config.toml, including a Stop hook |
learn.chatgpt.com |
| GitHub Actions | openai/codex-action, which installs the CLI and runs codex exec in a workflow |
learn.chatgpt.com |
| Headless mode | codex exec "<prompt>" |
learn.chatgpt.com |
| Rules files | AGENTS.md, AGENTS.override.md |
learn.chatgpt.com |
| Models | OpenAI's models, any compatible Responses API endpoint, and local open models through Ollama or LM Studio with --oss. Prices: OpenAI. |
learn.chatgpt.com, page 2 |
Each row was checked on 2026-10-17 against the maker's docs it cites. A feature with no row is one those docs don't describe.