How to Extract 73x Performance from Codex

"Codex is just a smart chat, right?"

If you are just throwing questions, copying the returned text, and throwing the next one, you aren't even using half of your Codex's potential.

Generating text. Creating posts. Mass-producing articles.

ChatGPT can do those things too.

Codex's true strength lies elsewhere.

How Codex Differs from a Simple Chat

The decisive difference between Codex and other chat AIs is that you can change its behavior with a single short command.

You can instruct the depth of thinking, the way it stores memory, and the replication of conversations all through commands.

Moreover, it reads files directly and rewrites them for you.

You can create automation tools, and there's no need to copy and paste the generated text back into your files.

In short, Codex is not just a chat partner; it's a partner that moves its hands on your behalf.

What I'm about to introduce are 10 commands to make that partner work for real.

Knowing or not knowing these will change the usability by several times, so please bookmark this and master them.

As a quick promotion, I am currently giving away 55 major benefits, including a Codex textbook, installation methods, and monetization strategies for free. You can receive them immediately from the link below if you haven't already.

↓

https://utage-system.com/line/open/cwgwX1a35XDK?mtid=FNAamIuYaEet

Now, let's get to the main topic.

1. /goal: Give it a goal and let it run on its own

A command where you set one completion condition, and Codex keeps running by itself until it reaches it.

Just write the goal you want it to reach after /goal.

Implement according to plan -> Test -> Review -> Re-implement.

It will repeat this automatically even if a human isn't standing by.

It's suitable for tasks where the direction is set, the volume is high, and the goal is clear.

For example:

• Implementation after a detailed plan is made
• Creating launch materials after a concept is solidified

It will proceed without stopping for hours or days without complaining, even while you are sleeping.

2. /clear: Reset a cluttered mind

Codex's performance drops as the conversation continues.

It becomes what people call "stupid."

This happens because the conversation history accumulates as context.

/clear solves this.

By using /clear when a task reaches a milestone, the context is reset, and a fresh, high-performance Codex is restored.

However, be careful as it will forget the context up to that point.

If you want to carry over necessary information, this prompt is recommended:

1Please save the contents decided so far into a document.
2After that, I will clear this session and start a new one, so please create a text to instruct what to do next and pbcopy it.

It will even copy the next instruction to your clipboard, so you can start working immediately by just pasting after /clear.

3. /personality: Change Codex's personality

A command to switch the character of Codex's replies—specifically the tone and energy.

When you type /personality, you can choose from prepared styles.

There are cases where you want it to return just the main points dryly, and cases where you want it to explain things more simply.

Instead of asking every time, you just choose once at the beginning.

Pragmatic will answer in a concise, task-oriented, and straightforward tone.

Friendly will answer in a warm, cooperative, and kind tone.

It's a subtle thing, but because it's a partner you work with side-by-side, whether the energy matches matters.

4. /review: Let Codex inspect what it created

A command to make Codex re-examine what it has made or fixed.

Things made by AI often have parts that differ from instructions if left as is.

Using /review allows Codex to inspect the production currently in your hands.

It will point out if there are deviations from the specifications in the plan or if there are missing parts.

5. /fork: Duplicate the conversation and try a different route

A command to copy the current conversation into a separate branch.

When you type /fork, a conversation starts in a new session while maintaining the flow up to that point.

"I'm torn between proceeding with Plan A or Plan B for the rest of this."

In such cases, you can try Plan A in one and Plan B in the other without messing up the original conversation.

It's like "Save As" for a conversation.

Even if you fail, the original remains, so you can give bold instructions.

6. /btw: A quick tangent without stopping the work

A command that allows you to ask a one-off question without polluting the main conversation thread.

Questions asked with /btw and their answers do not remain in the main context.

"By the way, what was that setting earlier?"

"What does this word mean again?"

Simple questions like these.

If you mix these into the main conversation, the context gets messy later.

You take a tangent, but the main thread remains 100% clean.

You can ask what you want to know and return quickly.

By the way, /side does the same thing.

7. /model: Switch the model and thinking depth on the fly

A command to switch which model you use and how deeply you want it to think.

Typing /model brings up a menu for you to choose from.

In addition to model selection, thinking depth has stages like low, medium, high, and Extra high.

For tasks where you want it to "think thoroughly," like investigating complex causes or complicated implementations, Extra high is recommended.

For writing, running it on high is recommended.

You use it like turning a dial according to the weight of the task.

8. /plan: Make it create a plan before starting

A mode to make it research and plan first instead of creating immediately.

If you write what you want to do after /plan, Codex will first provide the steps.

It's effective for complex implementations or large-scale corrections.

Moving after approving the plan drastically reduces rework.

"Starting to build for now, only to redo everything later."

This is the biggest waste. /plan exists to prevent that.

It's recommended to refine the plan with /plan and then let it work tirelessly with /goal.

9. AGENTS.md: Set rules to be read every time

A file like an instruction manual that Codex reads every time.

This is AGENTS.md.

If you write writing style rules or forbidden words here, it will read them every time across sessions. If you write rules you absolutely want followed or the directory structure, Codex can work without getting lost.

Once written, you never have to explain it again.

Initially, typing /init will make Codex look at the folder contents and create a draft automatically.

Then you just trim it to your liking.

Now, I'll tell you the most common mistake beforehand.

It's packing too much in.

If you overload it with rules, Codex won't know which to prioritize and will actually become inconsistent.

Keep only what would definitely cause a deviation if missing. That's just right.

10. /compact: Fold long conversations to stay light

A command to summarize and compress long conversations.

You can increase free capacity while maintaining the flow so far.

When Codex's replies become heavy in the middle of 20 or 30 exchanges.

Typing /compact makes it lighter while leaving the flow of the story.

By the way, you can see the remaining Context window capacity in numbers by typing /status.

In my experience, Codex's /compact feels like it compresses context while retaining memory better than Claude Code.

However, in reality, Codex judges automatically and uses /compact when the context is pressured, so you can leave it to the AI. You only need to use the command when you want to compress at your own timing.

Mastering commands multiplies Codex's usability

Honestly, if it's a short task, you don't need to distinguish this much.

But the longer and larger the task becomes, the difference opens up like heaven and earth.

To summarize:

• /goal: Let it run to the goal on its own
• /clear: Reset the mind
• /personality: Decide the tone
• /review: Inspect what was made
• /fork: Duplicate conversation to try different routes
• /btw: Protect the main thread even during tangents
• /model: Adjust model and thinking depth
• /plan: Plan before running
• AGENTS.md: Never explain the same thing twice
• /compact: Fold long conversations to stay light

You don't have to learn them all at once.

Bookmark this and try using them where you can.

I usually post about the latest AI information and monetization methods using AI. If this article was helpful, I'd be happy if you follow or quote it.

Also, I am currently giving away 55 major benefits, including a Codex textbook, installation methods, and monetization strategies for free. You can receive them immediately from the link below if you haven't already.

↓

https://utage-system.com/line/open/cwgwX1a35XDK?mtid=FNAamIuYaEet