[FEATURE] Model/Effort Selection at Gate Prompts

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

When Claude Code pauses at a gate prompt (e.g., "Here's the implementation plan — should I proceed?"), the only options are to approve, reject, or add notes. There's no way to adjust the model or effort level before execution begins.

This matters because the planning and execution phases of a task have fundamentally different compute needs. Planning — understanding requirements, mapping a codebase, designing an approach — benefits from a high-capability model at medium-to-high effort. But executing an already-approved plan (writing code that follows a clear spec) often doesn't need the same horsepower. Today, users either burn expensive compute on routine execution or manually toggle /fast and hope they remember to switch back.

Proposed Solution

Add a model/effort selector to the gate prompt UI. Before accepting a plan (or any gate prompt), let the user:

1. Pick the model (Opus, Sonnet, Haiku)
2. Pick the effort level (low, mid, high)
3. Then approve, approve with notes, or reject — as they do today

This keeps the current flow intact but gives users a natural decision point to right-size compute for the next phase. The selector could appear as a small inline option at the gate prompt, not a separate step.

Alternative Solutions

Add shorthand suffixes to the approval response, e.g.:

• "yes, execute with sonnet"
• "approve, switch to sonnet mid"
• Quick-action buttons like [Approve + Opus] [Approve + Sonnet] [Approve + Haiku]

This would be less intrusive to the UI but less discoverable for new users.

Priority

Medium - Would be very helpful

Feature Category

Interactive mode (TUI)

Use Case Example

Using Claude Code's plan mode:

1. User enters plan mode and asks Claude (Opus, high effort) to analyze a complex bug across multiple services. Claude investigates, reads files, traces the issue, and presents a detailed plan with the root cause and proposed fix across 4 files.

1. User reviews the plan and is satisfied — but the fix itself is straightforward: update a conditional, add a null check, rename a variable, and add a unit test. Before accepting the plan, the user selects Sonnet / mid effort and then approves. Claude exits plan mode and executes the fix efficiently without burning Opus-level compute on routine code changes.

1. After implementation, the user wants to verify the fix is correct. They switch back to Opus / low effort to ask Claude to review what it just wrote — analytical work, not generative. Opus confirms the fix is sound in a few sentences.

The pattern: Opus thinks, Sonnet builds, Opus verifies. The gate prompt between plan and execution is the natural place to make that switch.

Additional Context

• The gate prompt is the perfect natural decision point because the user is already pausing to review output and decide next steps
• This would help users who have plenty of credits but want to use compute responsibly — not every phase of a task needs the strongest model at full effort