Collaborative Page Tools for Codex Plugins

[leimenghao1991]

I would like to propose a new plugin capability for Codex: developer-defined interactive page tools that users and Codex can operate together.

Today, Codex plugins can extend Codex through skills, app integrations, MCP servers, hooks, and related metadata. This is powerful for tool access and workflows, but there is still a missing product surface: plugin-owned interactive pages where users can directly manipulate structured data, then ask Codex to help reason over, organize, edit, comment on, or act on that data.

This is not a request for Codex to continuously manage user state in the background. It is also not just asking Codex to visually inspect and click arbitrary UI. The idea is to let developers build interactive page tools with an agent-readable contract, so both the user and Codex can work over the same structured workspace.

Proposal

Add support for "Collaborative Page Tools" in Codex plugins.

A plugin could register a page UI, plus a structured API/state contract that tells Codex how to understand and operate that page.

Conceptually:

• Page UI: the human-facing interface, such as a task board, calendar, table, dashboard, editor, or PR queue.
• Page state API: exposes structured page data, current selection, filters, recent changes, and element metadata.
• Action API: exposes safe operations Codex can call, such as create, update, delete, classify, prioritize, summarize, or recommend.
• Element comments: lets users comment directly on a specific task, calendar event, table row, chart node, PR item, etc.
• Page-scoped composer: lets users ask natural language questions inside the page context.
• Approval model: mutating actions should show proposed changes and follow Codex's existing approval/safety model.

The key idea is:

The page is a shared workspace for the user and Codex. The user can directly interact with the UI, and Codex can read structured state, understand context, reply in place, and apply approved actions through a declared contract.

Example: Task Management Page Tool

Imagine a developer creates a task management plugin.

The plugin provides a task board page. The user can manually add simple tasks, drag tasks between columns, edit due dates, add notes, and comment on individual task cards.

The same plugin also declares a structured contract for Codex:

{
  "page": {
    "id": "tasks",
    "title": "Tasks",
    "entry": "./pages/tasks/index.html",
    "placement": ["sidebar", "main"]
  },
  "resources": {
    "task": {
      "fields": ["id", "title", "status", "priority", "dueDate", "project", "notes", "tags"]
    }
  },
  "actions": {
    "createTask": { "mutates": true },
    "updateTask": { "mutates": true },
    "classifyTasks": { "mutates": true },
    "rankTodayTasks": { "mutates": false },
    "summarizeTasks": { "mutates": false },
    "replyToElementComment": { "mutates": false }
  }
}

A user might manually add these tasks:

• Fix login timeout issue
• Prepare next week's release plan
• Reply to customer feedback about data export
• Review plugin page proposal

Then, in the page composer, the user types:

Classify the tasks I just added and recommend what I should work on today.

Codex can read the recently added tasks, classify them, rank them by urgency and impact, and show a proposed update:

• Bug: Fix login timeout issue
• Planning: Prepare next week's release plan
• Customer support: Reply to customer feedback about data export
• Product proposal: Review plugin page proposal

Recommended order for today:

1. Fix login timeout issue
   Reason: it may block users and QA.
2. Reply to customer feedback about data export
   Reason: an external stakeholder is waiting.
3. Prepare next week's release plan
   Reason: important, but less urgent.

The user can accept all changes, accept selected changes, or reject them.

Element-Level Comments

The user could also comment directly on a specific task card:

Should I prioritize this today?

Because the comment is attached to a structured element, Codex receives context like:

{
  "elementId": "task_123",
  "resource": "task",
  "comment": "Should I prioritize this today?",
  "context": {
    "title": "Fix login timeout issue",
    "status": "todo",
    "priority": "medium",
    "dueDate": "tomorrow",
    "project": "Mobile App"
  }
}

Codex can reply in place:

I recommend prioritizing this today because the due date is close and it may block QA. Suggested action: move it to High priority and schedule it for this morning.

The page could then show action buttons:

• Set high priority
• Schedule for today
• Split into subtasks
• Ignore

Why This Is Useful

This would make Codex plugins much more useful for structured workflows.

Many workflows are not ideal as pure chat:

• Task management
• Calendar planning
• PR review queues
• Incident dashboards
• Deployment panels
• Data tables
• Research boards
• Knowledge base organization

Users often need to see, filter, select, edit, and comment on structured objects. Chat is good for reasoning and commands, but pages are better for persistent structured interaction.

With Collaborative Page Tools:

• Users do not need to copy page data into chat.
• Codex does not need to guess the DOM or rely on visual automation.
• Developers can expose a clear, safe, structured contract for agent actions.
• AI responses can be attached to specific page elements instead of floating in a conversation.
• Users remain in control because they can directly manipulate the UI and approve mutating actions.
• Codex plugins can evolve from "tool bundles" into interactive, agent-aware workspaces.

[leimenghao1991]

This could become the missing protocol layer between agent tools and human-facing applications: a standard for shared workspaces where humans interact visually and agents operate structurally.