Command Center on the web
This evening I answered six decisions on a web page, and three of my notes said I needed more explanation. The page had no way to say that. It had options and a note box, and an answer closed the issue.
The page is the Command Center. The design says the markdown docs are scattered and the number of agents keeps growing, and it asks for one place that shows what is going on and what is waiting on me. I approved it on the 5th and added the decision inbox that same day, then Claude wrote a 12-task plan and built the first version on the 6th. A major decision, meaning one that blocks work or changes a design I already approved, becomes a GitHub issue with a decision label, with the options and a recommendation written out. I answer on the page, which posts my answer back on the issue so any session can read it.

The screenshots here come from a test copy with sample data. The first answer I gave on the real page went to issue #20 on the evening of the 6th, a test question from a fresh agent about adding a pointer line to another instruction file. I picked A, and a second fresh agent read the answer back correctly. I used the page the next day and sent feedback, mostly labels and links in place of sentences, plus an Agents page that draws the running sessions as a diagram. That was revision 2, and it merged on the 8th.
Hosting it
On the 11th I approved revision 3, a copy on Vercel. The goal I gave was one hosted URL that I always use, showing a change soon after I or any session updates the project. Claude built it in five steps across the 11th and 12th. The hosted page reads GitHub live, and each branch keeps a workstream file in git, which I chose over a database because a database would drift from the markdown docs. A small reporter on my machine writes the running sessions, agents and worktrees to a store that expires.

The reporter is the shaky part. On the 13th the Your move list held 16 items and none of them were real, because sessions wrote lines for me and nothing cleared them after a merge. I ruled that an item has to be actionable: “If there’s an open decision, it needs to use the GH issue decision framework + dedicated UI.” The Docs page now warns when an open engine decision has no issue. The hook that started the reporter also missed sessions that begin in another folder, so a tray icon starts it now.

Answering
Until today the hosted page could only read. Claude wrote up two options for letting it write an answer, and I picked B.
That went in around 7 pm. A little later I started on the first six of the 27 decision issues that the four epic briefs raised, the E1 ones about art and maps. On the map scales and the art style I asked for more info, and on the art style I wrote “I don’t really understand what this is asking. I need a deeper explanation and visual examples.”
So I asked for an Ask for more info button. Claude wrote the brief, and a verifier graded it three times before I approved it, 7.0 on the first round and a pass on the third. For the API I picked option A, which adds one action to the answer function that already exists. A second box, “Your question”, posts a comment that starts with Question: and adds a needs-info label, and the issue stays open. When a session starts it finds the question, answers with an Info: comment and takes the label off. The page then shows “More info posted.” Claude built it this evening, and it merged at 11:54 pm. The brief says to reopen my six decisions as waiting for info once it merges.
Tonight the page takes either an answer or a question. Most of the 27 decisions still wait for me.