Your agent didn't wait for you: the answer it assumed, and your OK

You start three agents and go to lunch. Ten minutes in, one of them reaches a decision that is yours: in a habit tracker, does a day end at midnight, or at 4am for the people who log their evening run at one in the morning? The agent has two bad choices. It can stop, and you come back to a terminal that did nothing for fifty minutes. Or it can decide alone, and you find out next week, in the code.

Hamloy gives it a third: assume an answer, say so where you will see it, and carry on. This post is that moment from start to end on Pebble, a small habit tracker, with the replies word for word.

The decision is on the list first

The agent doesn't ask in the chat, where a question is gone when the session closes. It adds a question to the project's lists, with the options and the one it recommends, and the task that depends on the answer waits on it. When an agent claims that task, the reply says what stands in the way and what it can do about it:

⏺ hamloy - Claim an item (MCP)(key: "B-lm4l")
  ⎿  B-lm4l@1 "Start a new day at the time the person picks" claimed by this session
     note: it still waits on Q-229x "When does a day end?"; do only what doesn't depend on that, or assume an answer and go on

It assumes, and says so

⏺ hamloy - Assume an answer (MCP)(key: "Q-229x", option: 2)
  ⎿  Q-229x@2   When does a day end?  options: 1) At midnight  2) At 4am, for night owls (recommended, assumed)

The question isn't answered. It stays open, on Questions, for you, with the option marked assumed, and the work that waited on it can go ahead.

The other agents on the project hear it under their next tool reply, so they know which answer the work is going ahead on:

Hamloy news (what changed since you last looked):
- claude (session 3f9a2c71) assumed an answer to Q-229x "When does a day end?": At 4am, for night owls; the human confirms or overturns it

Claude builds the 4am version and ticks B-lm4l.

You see it when you come back

Right now, the page that says who is working and on what, has a part for exactly this: Answered for you, needs your OK. Every question an agent went ahead on is there with the option it picked, like the one about reminders below, and the line under the page's title counts them.

You didn't have to be in any terminal when it happened, and you don't have to find the session that asked.

The same answer confirms it; another opens the work again

Pick the option the agent assumed and it is confirmed. Nothing else changes.

Pick another, and Hamloy does the part you would otherwise have to remember: every task ticked done on the assumption opens again, with a note to redo it for the real answer. Here the answer was told to an agent, which recorded it; on the board it is a click on the option.

⏺ hamloy - Record an answer (MCP)(key: "Q-229x", answer: "1", why: "Simpler to explain.")
  ⎿  Q-229x@3 answered: At midnight
     open again, built on the assumed answer "At 4am, for night owls": B-lm4l

That reply is what makes a guess safe to allow. Nobody has to keep in their head what was built on it.

What can't be assumed

A question with no options can't be assumed, because there is nothing to pick. The agent is told:

Q-q0ae has no options to assume; give it some (the edit tool with options), or leave it to the human

And only a question can be assumed. What only you can do, an account, a key, a payment, is a request: it waits for you, with its steps.

You keep the last word

Hamloy doesn't pick the answer and doesn't run the agent. It keeps the question where every agent and you can see it, marks what was assumed, and puts the work back when you say otherwise.

The details are in the docs: Questions, the assume tool and Human in the loop; what Right now lists is in The overview. To try it on your own repo, create a free board. It is free for one project.