[FEATURE] Per-task model & effort selection before execution

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

Right now the model and reasoning effort are fixed when a session starts. But within one session, tasks differ enormously in what they need: forging a principle into a canon file demands maximum reasoning; renaming a variable does not. Paying Opus-with-max for a one-line CSS fix is waste; paying Sonnet-with-low for an architectural decision is worse — it produces plausible instead of correct.

Proposed Solution

Proposal: let Claude assess the task before executing it and either (a) recommend a model/effort with a one-line reason the user can accept, or (b) switch automatically when the user has opted into that.

Key detail that makes it honest: the assessment is a recommendation with a stated reason, not a silent switch. "This touches the core principles file — suggesting max effort" is checkable. A silent downgrade is not.

Subagents already accept per-call model and effort, so the mechanism exists — what's missing is applying it to the main loop, and the pre-flight judgement.

Why it matters beyond cost: the mismatch runs both ways. Under-powering a hard task wastes more than it saves — the user pays for a wrong answer twice: once in tokens, once in the bug that ships.

Alternative Solutions

_No response_

Priority

High - Significant impact on productivity

Feature Category

Interactive mode (TUI)

Use Case Example

_No response_

Additional Context

_No response_