Skip to content
    ↑↓ select↵ openesc close
    English中文
    lars-hagen

    Openai Codex Headers

    v1.2.7Other
    opencode-openai-codex-headers

    Stopgap opencode plugin that smooths the OpenAI/Codex GPT-5.6 rough edges (Luna 404s, missing session titles, reasoning-summary markers) until native support lands

    Actively maintained
    #opencode#opencode-plugin#plugin#openai#codex#chatgpt
    GitHub reponpm pageHomepage
    GitHub stars

    0

    Monthly installs

    130

    20 in 7 days

    Composite scoreSCORE

    31.1

    Multi-signal model

    Last commit

    1 month ago

    2026-07-10

    Install and configure

    opencode.json

    Writes to this project's opencode.json — applies to this repository only.

    opencode.json

    {
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-openai-codex-headers@1.2.7"]
}

    opencode loads npm dependencies through its embedded runtime on startup and caches them locally — no manual global install needed.

    README.md
    View original→

    npm

    An opencode plugin for ChatGPT Plus/Pro users on the Codex backend. It smooths over the rough edges in opencode's OpenAI/Codex integration for the GPT-5.6 models (Sol, Terra, Luna) until native support lands upstream.

    What it fixes

    • Luna 404s. GPT-5.6 Luna models return 404 Model not found. The Codex backend gates them on the client identity, and opencode identifies as opencode rather than the Codex CLI. The plugin restores the genuine Codex signature so Luna is served, and keeps Terra in the Codex priority tier so it stops load-shedding with server_is_overloaded under contention.
    • Missing session names. opencode's title agent runs on a Luna model, so the same 404 leaves every session stuck on its default New session - <timestamp> name. Recovering Luna fixes the titles.
    • Reasoning-summary markers. GPT-5.6 ends each reasoning summary with an empty <!-- --> comment that opencode's TUI renders literally under the "Thought" headline. The plugin strips it on the wire so only the headline remains.

    Install

    Via the opencode CLI (-g writes it to your global config; drop it for the current project):

    opencode plugin opencode-openai-codex-headers -g

    Manual: add it to the plugin array in your opencode.json:

    {
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-openai-codex-headers"]
}

    How it works

    Identity. A chat.headers hook overwrites originator and User-Agent on the openai provider only, so requests present the genuine Codex CLI signature (the backend requires both). It loads after opencode's internal hooks and runs against the same shared output, so it wins; every other provider is untouched.

    Reasoning cleanup. opencode exposes no hook to transform reasoning text, so the plugin rewrites the summary events on the wire, on both the HTTP/SSE path and opencode's experimental WebSocket transport. It touches only reasoning delta events on /responses endpoints (host-agnostic, so it also works through a proxy or custom baseURL); everything else passes through unchanged.

    License

    MIT