Codex Sublime Text plug-in

Chat with the Codex CLI directly from Sublime Text. The plug-in spins up a codex proto subprocess, shows the conversation in a Markdown panel and lets you execute three simple commands from the Command Palette.

Features

• Full Codex capabilities
  • Assistant-to-Bash interaction
  • Sandboxing (on macOS and Linux)
  • Model and provider selection
• MCP support (via ~/.codex/config.toml)¹
• Deep Sublime Text integration
  • Multiline input field uses Markdown
  • Selected text is auto-copied into the message with syntax applied
  • Outputs to either the output panel or a separate tab
  • Symbol list included in answers
• Works out of the box[^2].

Installation

1. Install the Codex CLI (the plug-in talks to the CLI, it is not bundled).

npm i -g @openai/codex

On Windows you can use

winget install OpenAI.Codex

By default the plug-in uses “which(codex)”

If yours lives somewhere else, set the codex_path setting (see below). It accepts either a single string (e.g. "codex.exe" when Codex is installed natively on Windows) or an array of strings – e.g.:

{
  "codex_path": ["wsl", "-e", "codex"],
}

1. Plugin installation

   1. With Package Control Package Control: Install Package → Codex
   2. Manual Clone / download into your Packages folder (e.g. ~/Library/Application Support/Sublime Text/Packages/Codex).

2. Optionally create an OpenAI token and tell the plug-in about it.

Open the menu → Preferences › Package Settings › Codex and put your key into the generated Codex.sublime-settings file:

{
       "token": "sk-…"
   }

That’s it – hit ⌘⇧P / Ctrl ⇧ P, type Codex, select one of the commands and start chatting.

Commands (⌘⇧P)

• Codex: Prompt – open a small Markdown panel, type a prompt, hit Super+Enter.

• Codex: Open Transcript – open the conversation buffer in a normal tab.

• Codex: Reset Chat – stop the Codex subprocess, clear the transcript and invalidate the stored session_id so the next prompt starts a brand-new session.

Per-project configuration

Every Sublime project can override Codex settings under the usual settings section. Example:

{
    "folders": [{ "path": "." }],

    "settings": {
        "codex": {
            // will be filled automatically – delete or set null to reset
            "session_id": null,

            // model & provider options
            "model":            "o3",
            "provider_name":    "openai",
            "base_url":         "https://api.openai.com/v1",
            "wire_api":         "responses",
            "approval_policy":  "on-failure",

            // sandbox
            "sandbox_mode": "read-only",
            "permissions": [
                // additional writable paths (project folders are added automatically)
                "/Users/me/tmp-extra"
            ]
            ,
            // Auto-fold specific sections in the transcript by their header
            // name (case-insensitive). You can pass a string or a list.
            // Example: fold the model's internal reasoning block
            // (rendered as "## agent_reasoning").
            "fold_sections": ["agent_reasoning"]
        }
    }
}

Writable paths passed to Codex

The plug-in constructs the sandbox_policy.permissions list for each session:

1. /private/tmp
2. cwd – the first project folder (or the current working directory if there is none)
3. All folders listed in the Sublime project (visible in the sidebar)
4. Any extra paths you add via settings.codex.permissions

Those paths are sent to the CLI unchanged; Codex is free to read/write inside them depending on the selected sandbox_mode.

Default configuration sent to the CLI

The first thing the bridge does is send a configure_session message:

{
    "id": "<session_id>",
    "op": {
        "type": "configure_session",

        // model / provider
        "model":            "codex-mini-latest",
        "approval_policy":  "on-failure",
        "provider": {
            "name":     "openai",
            "base_url": "https://api.openai.com/v1",
            "wire_api": "responses",
            "env_key":  "OPENAI_API_KEY"
        },

        // sandbox
        "sandbox_policy": {
            "permissions": [
                "disk-full-read-access",
                "disk-write-cwd",
                "disk-write-platform-global-temp-folder",
                "disk-write-platform-user-temp-folder",
                {
                    "disk-write-folder": {"folder": "$HOME/.cache"} // for clangd cache
                }
            ],
            "mode": "workspace-write"
        },

        "cwd": "<cwd>"
    }
}

All values can be overridden per-project as shown above.

Enjoy hacking with Codex inside Sublime Text! 🚀

Code sent to the language model

The plugin only sends the code snippets that you explicitly type or select in the input panel to the language model. It never uploads your entire file, buffer, or project automatically. Local configuration (such as sandbox permissions or project folders) is used only by the CLI to enforce file I/O rules and is not included in the prompt context.

However keep in mind that since this plugin and tool it relays on is agentish, any data from within your sandbox area could be sent to a server.

Suppressing noisy events

If the Codex backend floods the transcript with incremental updates such as agent_reasoning_delta, add them to the suppress_events array in your project-specific codex settings:

{
  "suppress_events": ["agent_reasoning_delta"]
}

Auto-folding sections

You can tell the transcript to auto-fold certain sections by header name. The match is case-insensitive and can be configured globally or per-project.

• Global (Preferences ▸ Package Settings ▸ Codex ▸ Settings):

{
    // ... other settings ...
    "fold_sections": ["agent_reasoning"]
  }

• Per project (.sublime-project under settings.codex):

{
    "settings": {
      "codex": {
        "fold_sections": ["agent_reasoning"]
      }
    }
  }

Notes - Folding is scope-based and targets the Markdown meta.section for that header. Only the section body is folded, so the header line shows with an inline ellipsis (row style), e.g.: ## agent_reasoning .... - The fold is applied right after the section is appended. If your syntax definition delays section scopes, the plugin waits briefly to target the correct section.

[^2]: If both requirements are met (1) codex is installed and (2) token in settings is provided.