codex-bridge plugin
A plugin that lets Claude Code draw images with gpt-image-2 and hand bulk work to GPT-5 subagents, all on your ChatGPT plan
- 243 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Runs codex exec with access to your working directory, and codex-implementer writes changes into project files after you agree
- Uses your codex login session and spends ChatGPT plan quota, and with OPENAI_API_KEY set it moves to paid API calls
- Briefs and code fragments are sent to OpenAI through the Codex CLI
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/codex-bridgeDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Runs codex exec with access to your working directory, and codex-implementer writes changes into project files after you agree
- Uses your codex login session and spends ChatGPT plan quota, and with OPENAI_API_KEY set it moves to paid API calls
- Briefs and code fragments are sent to OpenAI through the Codex CLI
This is third-party code. Review the repository files before installing.
What it does
codex-bridge connects Claude Code to an installed Codex CLI. Through it Claude gains two abilities: it creates and edits images with gpt-image-2, and it hands part of the work to GPT-5 subagents. Both modes reuse the login you already made with codex login, so no OpenAI API key is needed and usage comes out of your ChatGPT plan quota. The plugin carries six skills: generate-image, edit-image, asset-set, ask-codex, codex-review and codex-delegate. They fire on plain requests such as "make an icon" or "review my changes before I commit". The asset-set skill builds favicon, app icon and OG card sets. When it generates a series, each new image receives an earlier one as a style reference, so palette and stroke weight carry over. Claude opens the finished PNG itself to check it and regenerates if it missed. There are five subagents: codex-artist, codex-reviewer, codex-debugger, codex-implementer and codex-second-opinion. The reviewer and the debugger only read code and verify each finding against the sources. Only codex-implementer may edit files, and only after you agree. Codex's intermediate output never enters Claude's context, which keeps just the brief and the result. The plugin also puts two commands on PATH: codex-imagegen for images and codex-run for any Codex task from a terminal.
Who it is for. For people who work in Claude Code, pay for ChatGPT Plus, Pro or Team, and want images and bulk edits without spending their Claude quota.
Good fit when
- You need a set of icons or favicons in one style: the plugin pulls your palette from tailwind.config and passes the first icon as a reference for the rest
- The same refactor must be repeated across dozens of files, such as renaming a hook repo-wide, and you are ready to review the resulting diff
- You want an independent review of your changes before a commit from a different model that checks each finding against the code
- A test fails only in parallel mode and you want GPT-5's second look at the cause
- A cover, mockup or hero image for a landing page should be made straight from a Claude Code session and checked afterwards
Not a fit when
- You have no ChatGPT Plus, Pro or Team plan: without one Codex runs only on an API key billed per token
- You need native transparency without an OpenAI API key: real alpha is only available through the Codex CLI fallback that uses that key
- The edit is two lines: handing it to Codex and reviewing the result costs more than the edit itself
- The task needs architecture decisions or depends on the conversation history: Codex cannot see your chat with Claude
- You work on Windows: the plugin was not tested there
Example request
Generate a hero image for the landing page and save it in assets/, then have Codex review my changes before I commitLimitations
- Requires an installed Codex CLI with codex login done and a ChatGPT Plus, Pro or Team plan. If OPENAI_API_KEY is set, Codex switches to API billing.
- Image turns burn plan quota roughly 3–5 times faster than text turns.
- One image takes 1–4 minutes, and a repo-wide task can run over 10 minutes.
- Transparent backgrounds take two steps: generation on a flat #00FF00 key, then removal with a helper script that Codex ships.
- Tested on macOS and Linux, not on Windows.
How to disable. Disable the plugin with /plugin in Claude Code or remove it from the same menu. If you installed it by cloning into ~/.claude/skills/codex-bridge, delete that folder.
FAQ
Does codex-bridge need an OpenAI API key?
No. The plugin calls codex exec, which runs under your codex login session. Usage comes out of your ChatGPT Plus, Pro or Team quota. OPENAI_API_KEY is only needed for native transparent backgrounds.
Can a Codex subagent change my files without asking?
Only codex-implementer edits files, and only after you agree and with a clean working tree so the changes land as a reviewable diff. The reviewer, the debugger and the second opinion are read-only.
Why did I get out-v2.png instead of out.png?
Codex refuses to overwrite existing files and sometimes saves under a new name. The codex-imagegen wrapper finds the file and prints the real path, so use that one.
What image sizes does codex-imagegen accept?
Longest edge up to 3840 px, both edges multiples of 16, aspect ratio at most 3:1, and 655,360 to 8,294,400 pixels in total. For favicons or a 1200×630 card, generate a large master and resample locally.
How does the plugin keep a style across a series of images?
Each new image is generated with an earlier one passed as a reference through the --ref option, up to four of them. Palette, stroke weight, glyph scale and margins carry over without being re-described.
Discussion
No questions or reviews yet.
Related
PluginsA Claude Code skill that hands implementation-heavy work to Codex CLI while Claude keeps planning, review and acceptance.
A Claude Code plugin that hands coding tasks to Codex agents, monitors them and verifies the results independently
A plugin that syncs Claude Code, Codex CLI and Antigravity on one project and lets them hand work to each other.
OpenAI's coding agent that runs locally in your terminal, with IDE and desktop versions
Claude Code supervises while Codex, opencode and pi work on your quota: delegated edits, independent review, model discussions
The /codex-workflows skill: Claude Code writes a script, dozens of Codex agents run it, and the run shows on a map
Comparison with similar tools
Cards that share the main search query, and how they differ.
README in short
The repository covers installation into Claude Code through a marketplace, six skills, five subagents and two shell commands. Image options, Codex sandbox modes and a table of common failures get their own sections. The page ends with limits on quota, speed and platforms.