Proposal / Discussion: Native structured input forms for Codex workflows

[galvn]

Hi. Thank you for building this beautiful piece. Here's a minor idea I want to discuss/share. I've have codex put it all together for me:

---

I have several recurring Codex workflows where the main task is not really writing a free-form prompt, but collecting structured data and sending it to GPT as JSON, for like updating my local knowledge base.

Examples from my own usage:

• Daily routine / health logs: repeated daily fields, currently entered manually.
• KitchenVitals inventory updates: structured pantry / food / cooking log updates.
• ProjectContentIntake: a more complex local HTML form that produces JSON for GPT.

Today, we probably have a workflow like this:

1. Build or open an external form.
2. Fill in structured fields.
3. Generate JSON.
4. Copy the JSON manually.
5. Paste it into the Codex prompt box.
6. Submit it to GPT.

This works, but the form system lives outside Codex. In practice, it often means spending tokens / credits repeatedly to generate small JS scripts, local HTML pages, or Node servers just to collect structured input that could conceptually be handled by a native form surface.

What made me think of this

Codex already has app-native interaction surfaces in some situations.

For example, when Codex asks for permission to run a command, the app shows an approval UI. In Plan mode, there is also an interaction style where Codex can ask a multiple-choice question and the user can answer through the app UI instead of typing a prompt manually.

I understand these may be different internal mechanisms, and I do not know the details of the Codex product architecture, tool-calling layer, or approval system.

But from a user perspective, the important observation is:

If Codex can already present app-native choices that guide what GPT should do next, that feels very close to the general idea of forms.

So I am wondering whether this could evolve into a more general native structured-input framework.

Desired direction

I would like Codex app to support richer structured input surfaces inside the app itself.

Conceptually, a task, skill, plugin, or workspace could define an input UI/schema. The user fills it in inside Codex, and Codex sends the resulting structured payload directly into the conversation or tool flow.

This would make Codex useful not only as a text prompt interface, but also as a structured task intake interface.

Important distinction

This is related to the existing app-native question / approval experiences, but I am not assuming they are the same API or should be implemented the same way.

The ask is broader:

• native structured forms
• repeated task templates
• JSON payload generation
• validation
• defaults / presets
• direct submission into the GPT interaction
• possible integration with skills, plugins, or workspace-specific workflows

In other words, I am not specifically asking for “make request_user_input more powerful” or “reuse the approval UI.” Those are just examples that made me realize the app already has some native interactive surfaces.

Why this matters

For recurring workflows, the prompt box is not always the right input surface. Many tasks have stable data shapes, and free-form prompting introduces avoidable friction and errors.

An integrated structured-input layer could:

• reduce copy/paste
• reduce repeated token / credit usage for generating ad-hoc local form tools
• make recurring workflows more reliable
• let Codex workflows feel more like task-specific tools
• avoid every user having to build and host separate local HTML forms for common structured-input tasks

Possible first version

A minimal version could be schema-driven:

• field types: text, textarea, number, date/time, checkbox, select, multi-select, list/repeater, file/path
• output format: JSON
• submit action: insert or send the JSON payload to the current Codex conversation
• optional workspace-provided templates

Longer term, this could become a plugin/skill-facing API for defining structured intake UIs.

Open discussion

I am posting this mainly as a discussion starter.

As a user, I can see the workflow need clearly, but I do not know enough about the Codex product direction, internal toolchain, or UI architecture to know what the right implementation should be.

Is this kind of native structured-input framework aligned with where Codex could go?

Are other users building similar external local forms just to generate JSON for GPT?

Would it make sense for Codex to provide a first-class way for workflows, skills, plugins, or workspaces to define these input surfaces?