UX: confirm scope upfront, not after work is already done

## Problem

Brain charges into substantive work and then asks "good idea, I can go?" *after* doing some amount of work. By then the user's quota is already spent, and any course correction means throwing away what was done. User feedback:

> i get questions right now like 'good idea, i can go?' after its done some amount of work already and then ill caveat like 'yea kinda good but xyz first'. and some of my precious quota gets used before i get to clarify rather than asking me upfront some questions.

## Proposed behavior

For anything with side effects (tool invocations, workflow runs, file creation, credentials usage), the brain should clarify scope *before* committing. That means:

- Surfacing unknowns ("what organism? which Galaxy? which history?") up front instead of guessing
- Proposing a sketch of the approach and getting a green light before executing
- Keeping clarification cheap -- one short exchange, not a full plan

Pure Q&A and low-stakes exploration should stay frictionless. The gate is for commitments, not for every message.

## Relationship to #16

Complementary but not a duplicate. #16 enforces plan creation before phase-structured tool execution. This issue is about the conversational layer one step earlier -- confirming scope and intent before even proposing a plan, so the plan (or the work) doesn't get committed to the wrong direction.

## Motivation

Balances planning vs. ad-hoc exploration -- the failure mode we want to kill is "agent spends quota charging off, user redirects, quota already gone."

No files, tests, or entry points are named. Start by reading this issue alongside #16 and tracing how the brain handles tool invocations, workflow runs, file creation, and credential use. Done means commitments pause for one short scope-clarification exchange, while pure Q&A and low-stakes exploration remain frictionless.

Written by the indexing model from the issue text.