Inbox and outputs¶

Outputs are what the agent produces. The inbox is where anything needing your judgement waits. Together they are the review half of the core loop.

Outputs¶

An output is a deliverable: the blog post, the competitor teardown, the email sequence, the pricing page copy.

One output per shareable thing. If a task produces three deliverables that different people would receive, that is three outputs. If it produces one document with three sections that are only useful together, that is one output. The test is whether you would send them separately.

Tasks start with no outputs. The agent creates them at the start of a run, in the order it plans to work, and fills them in as it goes. So an empty output is not an error, it is a promise about what is coming.

Editing¶

Outputs are editable documents and they save as you type. There is no save button and no "submit for review" step.

This matters more than it sounds. When a draft is 80% right, the fastest path is fixing the 20% yourself, not describing the fix to the agent and waiting. Edit it. Then, if the correction is one that will recur, write it into a rule so you do not make it again.

An output is a draft until you publish it, and publishing is a deliberate act. Nothing leaves your workspace because a run finished.

Once published, you can share an output:

by link, which anyone with the link can read, with no Questo account needed;
by email, sent to people you name;
into Slack, posted to a channel, if you have connected the Questo agent app. See Slack.

An agent can ask for permission to publish rather than publishing itself. That request shows up in your inbox as a card, alongside everything else waiting on you.

The inbox¶

The inbox is the single queue for everything that needs a decision from you. Proposals from the agent, plans from routines, weekly reports, deliverables waiting to be published, and notifications that a scheduled run has finished.

There is one queue rather than notifications scattered across the app, because a decision you have to hunt for is a decision that does not get made.

What is in a card¶

A card can hold several things at once. A daily mission plan, for instance, arrives as one card holding the day's win condition and several proposed tasks, each with a line saying why it is there.

That means you resolve a card piece by piece. You can approve two of the proposed tasks and reject the third. The card itself is only finished once nothing in it is still waiting.

What you can do¶

Approve applies the change.
Reject declines that piece and leaves the rest of the card alone.
Dismiss closes the whole card without applying anything.
Run appears on cards that suggest a piece of work rather than a change. It opens a fresh agent chat with the suggestion as the prompt, so you can steer before anything happens.
X, in the card header, discards the card permanently. This is the way to clear an information-only card such as a weekly report, and it is the only way to remove one.

Show resolved brings back everything you have already handled, so you can check what you approved last week without keeping it in your face all of this week.

What the agent asks about¶

The agent applies most of its work directly. It creates tasks, updates missions, writes outputs, sets up schedules and drafts inbox cards without stopping to ask, because a workspace that interrupts you for every write is a workspace you stop reading.

It asks first for exactly two categories:

Writes to your knowledge base. Creating or editing tenets, skills and rules. The knowledge base is what every future run reads from, so a silent edit there quietly changes all of them.
Deletions. Anything that removes something.

There is one refinement. If the thing being changed is already on your screen, the agent applies the change live and you watch it happen, rather than sending you a card about a document you are looking at. The exception to the exception is approving an inbox card itself, which always stays with you. A knowledge write becomes a card precisely so that a person decides, and an agent that could both write the card and approve it would be holding both ends of that gate.

Unattended background runs never get the benefit of any of this. With nobody at the screen there is nobody to ask, so a scheduled run cannot resolve inbox cards or approve plans at all. Its proposals wait for you.