Original video in Chinese.
Key Takeaway
- Codex’s automatic titles can become misleading: a conversation gets its title from the meaning of the first turn. Once later discussion changes direction, the title no longer fits, making old conversations hard to find as they pile up.
- Use a Hook for automatic renaming: have Codex rename a conversation based on its actual content after two turns. Starting at turn six, check again every two turns, with a four-turn cooldown to avoid frequent changes. Describe the prompt in plain language, and Codex will write the Hook itself.
- Tools should adapt to you, rather than you adapting to them: a Hook does more than rename a conversation. It lets the tool adjust automatically to your workflow, a key step toward turning Codex from a general-purpose product into a personal system.
Once you’ve used Codex for a while, you’re bound to run into this problem: so many conversations that finding one becomes a real chore!
Why is it so hard? When you create a new conversation, Codex automatically names it and gives it a title. But those titles really can be hard to make sense of.
Look at this long list of conversations I’ve had with Codex. My mind goes blank when I look at them now, because I have no idea what’s actually inside.
For example, in the conversation called “Test MCP connection,” what I was actually testing was the MCP connection to my own website. From the title alone, I’d think it was some other MCP. Later in that same conversation, we also worked on fixing membership verification, which has even less to do with the title.
This happens because Codex names a conversation based on the meaning of the user’s very first question. If the discussion later moves in a different direction, that name becomes very inaccurate.
So the best way to solve this is to have Codex automatically update the title based on what’s actually happening in the conversation.
How do you make it update automatically?
This involves a tool called a Hook. In plain language, when a certain trigger condition is met, it automatically takes a particular action.
Creating this Hook is easy. Just tell Codex what you need.
Look, I told Codex to rename the conversation based on its content after two turns. A few minutes later, it had written the Hook.
I tested it, and it works very well. In a real situation, the first turn might just be asking Codex to find some information. The second turn is where your real intention becomes clear. That’s why I designed it to wait two turns before renaming.
That raises another question: what if the direction changes again after two turns?
I updated the Hook. The second turn triggers the first title change. Starting at turn six, it checks every two turns, and after each successful update it waits through a four-turn cooldown, so it doesn’t rename things unnecessarily often.
This way, even as my Codex conversations keep piling up, I can find what I need at a glance.
I’ve posted the prompt for making this Hook in the newtype community on Knowledge Planet (Zhishi Xingqiu). You can get it there and adjust the trigger conditions to suit your needs. Just explain it in plain language. Codex understands.
I think what’s really interesting about a Hook is that it doesn’t just change a title. You’re starting to make the tool adapt to your workflow.
Tools should adapt to you, rather than you having to adapt to them. That’s always been my philosophy.
OK, that’s it for this episode. If you want to understand AI, reclaim your individual sovereignty, and find like-minded people, come join the newtype community. See you next time!