Urgent.News

the world's headlines, one feed

Editions

AI

Claude Code Can't Draw — This Plugin Borrows Your Codex CLI Login

TL;DR: "Claude can't draw" is a budget problem, not a capability problem. codex-bridge (MIT) is a Claude Code plugin that routes image generation (gpt-image-2) and five GPT-5 subagents through the Codex CLI login you already have — billing against your ChatGPT plan instead of a second API key. I have not tested or run it; this is a source-reading review. How it works. Two bash wrappers shell out…

Claude Code, a tool that allows users to generate code with natural language prompts, has a limitation: it cannot draw images. However, a plugin called codex-bridge may provide a solution. This plugin routes image generation and several GPT-5 subagents through the Codex CLI login, utilizing the user's existing ChatGPT plan for billing instead of a separate API key.

The plugin operates through two bash wrappers that interact with Codex CLI's non-interactive mode, using your existing codex login. Text tasks operate in a read-only sandbox, capturing only the final answer, while image tasks use a workspace-write sandbox to ensure the output file is saved. The exit codes for the plugin are 0 for success, 1 for precondition failure, 2 for failure, and 124 for timeout.

The key distinction between Codex and Claude Code is that Codex runs as a separate process, preventing intermediate outputs from entering Claude Code's context. This means that every brief must be self-contained, as Claude Code has no access to the conversation history. Consequently, every subagent is instructed to verify its work before reporting.

The codex-delegate rubric, found in the SKILL.md file, outlines when it is appropriate to delegate tasks to Codex. Generally, low-judgment tasks such as mechanical multi-file edits, bulk scaffolding, and exhaustive sweeps should be delegated, while high-judgment tasks like architecture calls, ambiguous requirements, and anything requiring conversation history should be handled directly by Claude Code.

However, there are some caveats to consider. Using the plugin incurs a cost in terms of time and chat quota, as images take 1-4 minutes to generate and burn through the ChatGPT quota three to five times faster than text. Additionally, the plugin does not support native alpha channels, and it is currently limited to macOS and Linux systems.

Furthermore, the default settings only allow image sizes with a longest edge of 3840 pixels and a resolution of 655,360–8,294,400 pixels, with both edges needing to be multiples of 16. Lastly, the plugin cannot emit images and text simultaneously.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — it may contain errors, so check the original before relying on it.

Read the original at dev.to →

More in AI