Skills in repo not found in v0.88

I meet the same problem. In v0.88 cannot find custom skills in repo.

Same issue

If you created the skills after starting Codex, you'll need to restart the CLI (or extension) for the skills to be detected. They are enumerated only at startup time.

@etraut-openai Same problem, i restarted codex.

I'm not able to repro the problem. I can create new repo-specific skills, and they are detected by Codex when I rerun it. Are you sure you're storing the skill in the correct location? For details, refer to this documentation.

Using https://github.com/nextlevelbuilder/ui-ux-pro-max-skill and followed their tutorial exctly but codex doesnt see it.

````
+---.git
+---.codex
¦ +---skills
¦ +---ui-ux-pro-max
¦ ¦ SKILL.md
¦ ¦
¦ +---data
¦ ¦ ¦ charts.csv
¦ ¦ ¦ colors.csv
¦ ¦ ¦ icons.csv
¦ ¦ ¦ landing.csv
¦ ¦ ¦ products.csv
¦ ¦ ¦ prompts.csv
¦ ¦ ¦ styles.csv
¦ ¦ ¦ typography.csv
¦ ¦ ¦ ui-reasoning.csv
¦ ¦ ¦ ux-guidelines.csv
¦ ¦ ¦
¦ ¦ +---stacks
¦ ¦ flutter.csv
¦ ¦ html-tailwind.csv
¦ ¦ nextjs.csv
¦ ¦ nuxt-ui.csv
¦ ¦ nuxtjs.csv
¦ ¦ react-native.csv
¦ ¦ react.csv
¦ ¦ shadcn.csv
¦ ¦ svelte.csv
¦ ¦ swiftui.csv
¦ ¦ vue.csv
¦ ¦
¦ +---scripts
¦ core.py
¦ design_system.py
¦ search.py

````

@etraut-openai can you try reproducing it using that repo maybe?

Same problem in VS Code plugin, reverting back to version 0.4.60 solved it for me. I tried pasting the exact same skills that are recognized in my Personal section into the repo and nothing.

The CODEX development team doesn't seem to consist of actual competent humans; they're like a bunch of people just coasting along, doing nothing productive. CODEX always has serious bugs, is perpetually unstable, and will never be able to compete with Gemini CLI or Claude Code.

I’ve run into the exact same issue—my custom Skill simply can’t be viewed at all. I’m absolutely sure there are no configuration errors, and even GPT5.2 Thinking can’t resolve this problem for me.

I also encountered a similar problem.

As a workaround, in my environment I was able to change the trust_level of the target project in ~/.codex/config.toml and it became visible.

[projects."/path/to/project"]
-trust_level = "untrusted"
+trust_level = "trusted"

Since this is a security setting, it is best to be careful when changing it.

I also encountered a similar problem. As a workaround, in my environment I was able to change the trust_level of the target project in ~/.codex/config.toml and it became visible. [projects."/path/to/project"] -trust_level = "untrusted" +trust_level = "trusted" Since this is a security setting, it is best to be careful when changing it.

Nice find! But security isnt a concern when always being in `--yolo` mode xD
Codex could ask if it should trust the skill instead of silently ignoring it

Starting with version 0.88 of the CLI, we filtered local skills for non-trusted projects. Based on user feedback and additional risk analysis, we've decided to reverse that decision. The next version will once again enumerate local skills even for untrusted projects. Of course, skill execution will still be subject to the sandbox mode you choose.

@etraut-openai I think it would be very useful if the initial status box indicated the trust level, in some way.

The trust level isn't very discoverable today, so if anything is based on that trust there is almost no way to realize the gap.

So something like this:

╭─────────────────────────────────────────────╮
│ >_ OpenAI Codex (v0.88.0)                   │
│                                             │
│ model:     gpt-5.2-codex   /model to change │
│ directory: ~/workspace/project (untrusted)  │
╰─────────────────────────────────────────────╯

@corrylc, please open a new feature request if you'd like.

<img width="788" height="408" alt="Image" src="https://github.com/user-attachments/assets/bc4a3c46-bf1e-4eae-bedd-f2516170219c" />

I tried configuring the relevant settings, and it still didn’t work after I restarted the extension. What’s more, I’ve noticed that my Codex version has changed to 0.4.62.

I’m using the extension in VSCode.