Codex Desktop: saved task presets combining prompt, model, reasoning, permissions, and skills

What variant of Codex are you using?

Codex Desktop

What feature would you like to see?

Summary

Please add first-class saved task presets that bundle a reusable prompt template with the Codex settings needed to run it, and make those presets quickly selectable from the composer, command palette, slash menu, or an assignable keyboard shortcut.

Problem

Recurring workflows usually require changing several independent controls before entering nearly the same prompt:

• select a model;
• select reasoning effort;
• choose permissions / sandbox or work mode;
• activate one or more skills;
• enter a repeated prompt or instruction template.

Codex has mechanisms for many of these pieces separately, but there is no single user-facing preset that applies them together. The repeated setup adds friction and makes it easy to run a workflow with the wrong model or permissions.

Proposed behavior

Allow users to create named presets such as:

• Quick fix — fast model, low/medium reasoning, workspace permissions, prompt: "Diagnose and implement the smallest verified fix."
• Deep review — high-capability model, high reasoning, read-only/review mode, prompt: "Review the current changes for bugs, regressions, and missing tests."
• Plan only — high reasoning, plan mode, read-only permissions, selected planning skills.
• Commit — fast model, low reasoning, repository permissions, a reusable commit workflow prompt.

A preset should be able to include:

• prompt text, with optional placeholders/arguments;
• model;
• reasoning effort;
• permissions / sandbox policy;
• collaboration or work mode;
• skills or user commands to invoke;
• optionally other safe task-scoped settings.

UX

• Show saved presets in the composer and command palette.
• Support invocation through a slash command such as /preset deep-review.
• Allow optional user-defined keyboard shortcuts.
• Let users choose whether a preset inserts editable prompt text or sends immediately.
• Clearly preview the settings that will change before sending.
• Keep the current draft intact while selecting a preset.
• Show the active preset/settings in the composer.
• If a model or capability is unavailable, show an explicit error instead of silently falling back.
• For security-sensitive changes, do not silently increase permissions; require a clear confirmation or restrict application to new tasks.

Scope / distinction

This is not only a model preset and not only a saved prompt. The main value is applying the prompt and execution settings atomically as one reusable task workflow.

An MVP could support new tasks first, then optionally support task-scoped or one-shot presets inside an existing conversation.

Related issues

• #35340 — saved model/reasoning presets with assignable hotkeys
• #39437 — saved prompt favorites in the Desktop composer
• #39527 — Desktop selector for named config profiles
• #32961 — skills or user commands declaring model/reasoning
• #30700 — binding arbitrary prompt text to a keyboard shortcut
• #17545 — main-session role/agent presets

These requests cover individual parts of the workflow. This issue proposes a lightweight user-facing preset that composes them for recurring tasks.

Additional information

A useful implementation could store presets in a human-readable user or project configuration so they can be versioned and shared, while still exposing creation and editing through Desktop Settings.

No implementation files or tests are identified. Start by locating the Codex Desktop composer, command palette, slash menu, settings, and configuration entry points, then review the related issues listed in the proposal. Done means a named preset can atomically apply its prompt and execution settings with the stated preview, confirmation, availability, and draft-preservation behavior.