Codex: The Ultimate Painless Getting Started Guide

This tutorial is for beginners and those with partial knowledge. It is a spoon-fed guide with many illustrations, telling you everything you need to quickly get Codex up and running!\n\nCodex is an AI agent for programming and computer tasks. Simply put, it doesn't just "answer code questions"; it can read files in a project, modify code, run commands, fix bugs, write tests, perform code reviews, and even help you complete repetitive tasks across different tools. OpenAI officially describes it as an "AI agent that can be delegated real work."\n\nWithout further ado, let's look at how to use it:\n\n# Conversation\n\n!Image\n\n## Create New Project/Open Existing Project\n\nIn this part of the sidebar (as shown), there is a folder-like icon. Clicking it reveals two options:\n\n!Image\n\n- Create New Blank Project\n\nThis creates a New Project in the default Codex path. We generally do not recommend this approach.\n\n- Use Existing Folder\n\nWe recommend this one. Not only can you open existing projects, but for example, if you want Codex to access your PPT-related projects, you can just open your PPT folder here. Additionally, if you want to start new work in a blank project, it is recommended to create a folder in your system file explorer first and then open it with Codex. This avoids complications like being unable to find your project.\n\n## Model and Speed\n\nAfter entering a project, you can select the model, reasoning level, and speed in the chat bar (as shown):\n\n!Image\n\nEveryone should understand the models, so I won't say much.\n\nReasoning level refers to how much reasoning/planning capability Codex is willing to spend when completing a task.\n\nSpeed refers to the generation speed (High Speed currently consumes 1.5x).\n\nI recommend using GPT-5.5 + Ultra High + Default Speed for any task.\n\nThis combination is the most stable, and the consumption is acceptable. You can certainly execute simple tasks at a lower level based on the scenario, but the act of "judging which level to use" itself consumes your energy, and the instability of lower levels might lead to rework. Therefore, I believe keeping the GPT-5.5 + Ultra High + Default Speed combination unchanged is best.\n\n## Permissions\n\nClick here in the chat bar (as shown) to control Codex's permission boundaries:\n\n!Image\n\nThese three are Codex's Permission/Approval Modes. The main differences are: whether it can automatically modify files, whether it can automatically execute commands, and when it needs to ask you.\n\nDefault permissions are conservative. Codex needs to confirm with you before key operations, such as modifying files or running certain commands. It's safe but will pop up confirmations frequently.\n\nAuto-review is the middle ground. When Codex encounters an operation requiring confirmation, "Auto-review" first judges if it's safe; safe ones are automatically allowed, while uncertain/dangerous ones will ask you. It's more convenient than Default, but automatic judgment isn't absolutely reliable.\n\nFull Access is the most open. It tries not to block Codex, allowing it to read, write, and run commands more freely, potentially involving a wider range of operations.\n\nMy advice:\n- Just starting with Codex: Choose "Auto-review"\n- Handling important projects/company code: Choose "Default" or "Auto-review"\n- Only use "Full Access" in temporary demos, test repositories, or projects that can be restored at any time\n- Do not enable Full Access in directories containing keys, private files, or production environment configurations\n\n## Plan Mode\n\n!Image\n\nPlan Mode allows Codex to think through a solution before modifying code.\n\nWhen enabled, it usually does the following first:\n1. Read project context: Look at files in the workspace, project structure, and relevant code.\n2. Understand your goal: Confirm what you actually want to achieve.\n3. Propose an execution plan: Tell you which files it plans to change, the steps involved, and potential risks.\n4. Ask questions if necessary: If requirements are unclear or if it affects databases/interfaces/dependencies, it will confirm first.\n5. Implement after the plan is clear: Avoids messy changes from the start.\n\nWhen is it suitable?\n- Modifying multiple files\n- Adding new features\n- Refactoring a project\n- The cause of a bug is not obvious\n- You are unsure how to proceed\n- The project is important and you don't want it to make random changes\n\nSimply put: do small things directly, use Plan Mode for big things.\n\n## Review\n\n!Image\n\nThis shows the changes and outputs of each conversation, allowing you to review or undo them directly.\n\nUsually, you don't need to worry about this unless there's something specific to review; if there's an issue, just ask Codex to change it.\n\n!Image\n\nThis shows some environment information. It's recommended to configure the GitHub CLI locally. Developers can also configure CLIs for Railway/Vercel, Sentry, etc.\n\nThis will give you a much smoother experience.\n\n## Some Tips\n\n!Image\n\nYou can pin your frequently used or unfinished conversation windows here for quick access next time.\n\n!Image\n\nVoice input is very convenient here. Half of my prompts are now spoken.\n\n!Image\n\nYou can open the sidebar or bottom window here to access files, the terminal, or websites.\n\nWe can let it modify code while opening a browser here to see the results. Or open a prompt reference book to follow step-by-step. You can even put up a PPT or watch a movie while working.\n\n!Image\n\nYou can open the project folder here, making it very easy to find things.\n\n# Skills and Plugins\n\nPlugins = Installing capabilities.\nSkills = Specific ways to use those capabilities.\n\n## Plugins\n\n!Image\n\nClick the location shown to open plugins. Explaining this in detail is complex, so just know that Plugins are like installation/extension packages. Once installed, Codex gains certain abilities, such as Chrome, Spreadsheets, Presentations, Slack, etc.\n\nRecommended: Chrome, Presentations, SpreadSheet, Documents.\n\nChoose others based on your development needs.\n\nSelect them explicitly here when using them:\n\n!Image\n\n## Skills\n\nSkills are specific workflows or instruction manuals configured within plugins or locally. They tell Codex what steps to take, which scripts to use, and which templates to refer to when encountering certain tasks.\n\nWe can get them from GitHub, install them from Codex recommendations, or create our own skills.\n\nOn GitHub, you can usually do this: (Paste GitHub link), help me pull this skill and register it as a global skill.\n\nTo make your own skill, use this entry (as shown). After clicking create, it will give you a prompt where you can add your requirements.\n\n!Image\n\nThere are several ways to use skills. The first is implicit triggering based on skill settings—if your skill says to use it in situation XX, and that situation occurs, Codex will say: "OK, time to use XX skill."\n\nThere is also explicit use. After typing "$" in the input box:\n\n!Image\n\nYour skill list will appear. Find and click or type the first few letters to match, telling Codex: "I want to use this skill now!"\n\nI want to emphasize this skill: image gen!!\n\nIt seamlessly integrates gpt-image-2 capabilities into Codex. This means we can let Codex look at local design specs or content, plan a prompt, and generate images directly inside Codex.\n\nIt can also help you use the images immediately, such as putting them into code or inserting them into an article, greatly reducing friction!\n\n## Conclusion\n\nCodex has many other features like automation and work trees, which I won't go into here. I hope everyone explores them!\n\nI also hope this article helps you!!","description":"Unlock the full potential of Codex with this spoon-fed guide to using the powerful AI agent for coding, automation, and project management.","summary":"This comprehensive tutorial explains how to set up Codex, manage permissions, use Plan Mode for complex tasks, and leverage plugins and skills to automate your workflow.","political":false,"sexual":false,"moderationReason":"","slug":"codex-ai-agent-setup-guide"}```