M
Melvynx
#Codex#AI coding#developer tools

Codex AI Developer Updates: Appshots, Goals, and Mac Control

Explore Codex AI's latest features including Appshots for instant context sharing, the /goal command for autonomous task execution, and Computer Use for Mac automation. These updates streamline AI-assisted development workflows.

5 min readAI Guide

Introduction

Codex, an AI coding assistant, introduces new features to enhance developer workflow. These include "Appshots" for instant context sharing from any application, an integrated /goal command for autonomous task execution, and an in-app browser with annotation capabilities for interactive development and debugging. Additionally, Codex now offers "Computer Use" to securely manage tasks on a locked Mac, expanding its automation potential.

Configuration Checklist

Element Version / Link
Language / Runtime Not explicitly mentioned, but implies standard development environments (e.g., JavaScript/TypeScript for web apps, macOS for system interactions).
Main library Codex application (OpenAI)
Required APIs OpenAI API (for Codex functionality)
Keys / credentials needed User's macOS password (for "Locked Computer Use" plugin installation)

Step-by-Step Guide

Step 1 — Using Appshots for Context Sharing

Step 1 — Using Appshots for Context Sharing
Appshots allow you to instantly capture a screenshot and extract text from your current application window, including scrolled content, and attach it to your Codex chat. This provides immediate context for the AI, streamlining discussions about UI, code, or any on-screen information.

To use Appshots:

  1. Enable Appshots: When in a new Codex chat, a prompt will appear to enable Appshots. Click "Enable".
    • Alternatively, you can trigger it with a keyboard shortcut.
  2. Capture an Appshot: Press Command-Command (double-tap the Command key on Mac).
    • This will take a screenshot of your current active application window and extract its visible and scrolled text content.
  3. View in Codex: The captured Appshot will be instantly added to your current Codex chat thread.
    • Codex will also attempt to identify the application from which the Appshot was taken.

Example Usage:

// After pressing Command-Command while on a web page:
// Codex will add an image and text from the web page.
// You can then ask questions about the content:
quel application j'ai screenshot ici ? // What application did I screenshot here?
// Codex will respond with:
// Tu as screenshot 3 applications:
// 1. Notes sur macOS
// 2. Spark Mail
// 3. Helium / Chrome, ouvert sur Thumfa.st

Step 2 — Setting and Managing Goals with /goal

The /goal command allows you to define a specific objective for Codex to pursue autonomously. Once a goal is set, Codex will continuously work towards achieving it, launching agents and performing actions as needed, even across hours or days.

To use /goal:

  1. Initiate a Goal: In a new chat, type /goal and press Enter.
    • This will add a "Goal" button to your chat interface.
  2. Define the Goal: You can either type your goal directly into the chat, or click the "Goal" button to open an edit interface.
    • Manual Goal Creation (less recommended):
      /goal <intention> <condition> <rules>
      // Example:
      create a goal to verify that vercel is green and all ci is green and then do it
      
    • AI-Assisted Goal Creation (recommended):
      // Simply describe your desired outcome, and Codex will format it as a goal.
      create a goal to verify that vercel is green and all ci is green and then do it
      // Codex will then create a more structured goal:
      // Paused goal: Verify that the current Vercel deployment is green and all CI checks are green for /Users/melvynx/Developer/saas/thumfa.st
      
  3. Manage the Goal:
    • Edit Goal: Click the pencil icon next to the goal to modify its description or parameters.
    • Pause/Resume Goal: Click the pause/play icon to temporarily halt or restart the goal's execution.
    • Clear Goal: Click the trash can icon to delete the goal.

Step 3 — Iterating with the In-App Browser and Annotations

Step 3 — Iterating with the In-App Browser and Annotations
The in-app browser allows for direct interaction with web pages within Codex, providing a powerful environment for debugging, design iteration, and testing. Its advanced annotation mode enables precise feedback and instant preview of changes.

To use the In-App Browser:

  1. Open the Browser: In your Codex chat, click the "Browser" button in the right-hand sidebar.
  2. Select a URL: Codex will detect locally running web servers or previously opened URLs. You can select one or enter a new URL.
  3. Interact and Annotate:
    • Screenshot: Use the screenshot button (camera icon) to capture the current view.
    • Annotate Elements: Click the annotation button (pen icon) to enter annotation mode.
      • Click on any element on the page to select it.
      • A pop-up will appear allowing you to add comments or directly modify CSS properties (e.g., font-size, border-radius, color).
      • Apply changes to see instant visual feedback.

Example Annotation:

// In annotation mode, select an H2 title element.
// In the pop-up, change the font size:
font-size: 44px;
// Click "Apply".
// Codex will record the annotation:
// 1 annotation
// apply
// font-size: 30px -> 80px

Step 4 — Enabling Computer Use for Locked Mac Automation

Step 4 — Enabling Computer Use for Locked Mac Automation
The "Computer Use" plugin allows Codex to securely interact with and control applications on your Mac, even when the screen is off or locked. This enables continuous background automation for tasks that require desktop app interaction.

To enable Computer Use:

  1. Install the Plugin:
    • Go to the "Plugins" section in Codex.
    • Search for "Computer Use".
    • Click "Install Computer Use".
    • Toggle "Locked Computer Use" to ON.
    • Enter your macOS password when prompted to allow the installation of an Apple authorization plugin.
  2. Restart Codex: If the plugin doesn't appear or function correctly, quit and relaunch Codex.
  3. Use Computer Use: Once installed and enabled, you can instruct Codex to perform tasks involving desktop applications.

Example Usage (intended):

// In a Codex chat, with Computer Use enabled:
// (Assuming Mac is locked)
Utilise l'ordinateur, ouvre les notes Apple et écris "Codex" dans une nouvelle note.
// Use the computer, open Apple Notes and write "Codex" in a new note.

Safety Guidance (from OpenAI documentation):

  • Give Codex one clear target app or flow at a time.
  • You can stop the task or take over your computer at any time.
  • Keep sensitive apps closed unless they're required for the task.
  • Avoid tasks that require secrets unless you're present and can approve each step.
  • Review app permission prompts before allowing Codex to use an app.
  • Use "Always allow" only for apps you trust Codex to use automatically in future tasks.
  • Stay present for account, security, privacy, network, payment, or credential-related settings.
  • Cancel the task if Codex starts interacting with the wrong window.

Comparison Tables

Codex vs. Claude (Uptime)

Service Uptime (Last 90 days) Notes
Claude.ai 89.92% Experiences frequent downtimes, often not fully operational.
Claude Console 89.97% Similar uptime to Claude.ai.
Claude API 89.94% Similar uptime to Claude.ai.
Claude Code 89.91% Similar uptime to Claude.ai.
OpenAI APIs (overall) 99.98% Generally fully operational, with high uptime across components.
OpenAI ChatGPT 99.97% High uptime.
OpenAI Codex 99.80% High uptime, but some minor incidents.

⚠️ Common Mistakes & Pitfalls

  1. Appshots not working as expected: If Appshots don't capture scrolled content or text accurately, ensure the application window is fully active and visible before pressing Command-Command.
  2. Incorrect /goal formatting: When manually defining a goal, ensure you follow the /<intention> <condition> <rules> format. Otherwise, let Codex generate the goal from a natural language prompt for better results.
  3. In-App Browser refresh issues: The in-app browser might refresh frequently if the underlying code is being actively refactored or if there are errors, leading to a suboptimal user experience for live development.
  4. Computer Use plugin not appearing/working: If the "Computer Use" plugin doesn't show up or fails to function, try quitting and relaunching Codex. Ensure you've entered your macOS password correctly during installation to grant necessary permissions.
  5. Computer Use limitations with specific apps: Codex's "Computer Use" might struggle to interact with certain native macOS applications (e.g., Apple Notes) or specific UI elements, requiring manual intervention or alternative scripting.

Glossary

Appshots: A Codex feature that captures a screenshot and extracts text (including scrolled content) from the current application window, attaching it to a chat for context.
/goal: A Codex command that initiates an autonomous task, allowing the AI to continuously work towards a defined objective over time.
In-App Browser: A browser integrated within the Codex application, enabling direct interaction with web pages, annotation of elements, and instant preview of changes.
Computer Use: A Codex plugin that allows the AI to control and interact with applications on a macOS computer, even when the screen is locked, by installing an Apple authorization plugin.
Skills (Codex): Pre-configured or custom-created functionalities within Codex that allow it to perform specific tasks, often involving external tools or complex workflows.

Key Takeaways

  • Appshots significantly reduce friction in sharing context from any application directly into Codex chats, improving communication and AI understanding.
  • The /goal feature empowers Codex to autonomously pursue complex tasks over extended periods, acting as a persistent AI assistant.
  • The in-app browser with annotation capabilities offers a powerful environment for interactive development, debugging, and design feedback directly within Codex.
  • "Computer Use" extends Codex's capabilities to control desktop applications on a locked Mac, enabling true background automation for development workflows.
  • OpenAI's API and services generally demonstrate high uptime, providing a reliable platform for AI-powered development.
  • Custom skills and configurations are crucial for maximizing Codex's utility, allowing users to tailor the AI's behavior to specific project needs.
  • While powerful, features like "Computer Use" are still in beta and may have limitations or require specific workarounds for certain applications.
  • The continuous integration of AI into the developer workflow, as demonstrated by these Codex updates, aims to enhance efficiency and reduce manual effort.

Resources