Your Workbench just learned to ask you the right question.

A few sharpenings I've shipped to my own Workbench, packaged for you to add to yours. The headline is a new surface called the Decision Bench, plus a cleaner way to run your todos. You're already running the Workbench on localhost, so this isn't a reinstall. It's a set of improvements your Claude drops into what you already have, one at a time.

The idea

Two surfaces, one question each

The Workbench splits your work by the question it's asking you. Nothing lives in both places, so you always know which hat you're wearing.

Home asks: what should I do next?

Every card is a task, something that can be done, by you or by a Claude session you fire off. It's built from each project's own todo list.

Decision Bench asks: what needs me to decide?

Every card is a call only you can make: a yes or no on a direction, or a fact that lives only in your head. Not work. A decision.

The one rule that sorts everything: could Claude just do it? Then it's a todo on Home, never a decision. Does it need your judgement, your permission, or a fact only you hold? Then it's a card on the Decision Bench.

The big one

The Decision Bench

One screen you open to make the calls holding your projects up. A morning pass reads every project and asks Claude for the questions that actually need you, so the decisions come to you instead of hiding in your todo list.

n-gram automation

Should the Monday sweep apply the negatives itself, or keep proposing a list you approve?

The rebuild is ready. The only open question is whether it writes negatives or just surfaces them.

My takePropose-only for the first month, then revisit.

YesNo
pmax playbook

How should the playbook handle brand traffic in Performance Max?

The last restructure forced the call, and three more clients are queued behind the same question.

My takeSplit brand into its own Search campaign.

Split into SearchCap the budget
nownextlater   The left edge is the project's priority, the same colours as Home, so the whole Workbench reads as one thing.
  1. You answer, and it comes off the bench. Answered decisions never come back.
  2. The call is written to the project's notes. The reason a project changed direction is on the record, not just in a chat that scrolls away.
  3. A Claude session picks it up. Your answer doesn't edit the work itself, it starts a session that moves the project forward with your decision in hand.
Also in the box

Your todos, easier to run

The Home cards got a set of small fixes that add up to a board you can reshape in a few seconds.

Reorder inside a card. Drag your todos into the order you'll do them, and the new order saves straight back to the project's todo file.
Promote from the backlog. Drag a backlog item up onto the active list when it's time to work on it.
Delete a single todo. A hover × removes one todo, without opening a file or firing a session.
Open a todo in place. Click any todo for a modal with its full context, the files a session would read first, and a one-tap way to fire it off.

The cards themselves got tidier too: the actions moved to quiet icons that appear on hover, and a redo button re-runs the project's todo refresh.

Adding it to yours

How you get it in

This ships as a single zip, not a new skill, because your Workbench is your own and I don't want to overwrite what you've built. Inside is a START-HERE prompt written for your Claude, the reference code for each feature, and a small runnable preview.

  1. Unzip it and open the folder in your brain. Everything it needs is inside.
  2. Tell your Claude to read START-HERE.md and add the features you want. It reads the reference for each one and grafts it into your Workbench, adapting to how yours is laid out and testing as it goes.
  3. Run the preview first if you'd like to see it working before it touches your own board. One command, a sample workbench, all the new features live.

Take the features you want and leave the rest. Nothing here assumes your Workbench looks exactly like mine.