Connected Done to your AI — now what? Five prompts to try first

Once Done is connected, you don't open a new screen — you just ask your assistant, in plain language, from the chat you already use. Five prompts cover most of what people want on day one: add a task, import a list, break a big task down, work the queue highest priority first, and put that on a schedule. Each one maps to a handful of Done's MCP tools, and none of them lets the agent close a task you haven't trusted it to close: unfinished work it hands back lands in Needs you for your approval, not in Done.

1. How do I add a task by just asking?

Try: “Add “renew my passport” to my Done board, due next Friday.” The assistant calls create_task with a title and a due_date, or quick_capture, which reads cues like “next friday”, “at 3pm”, “p1” and “#project” straight out of the text. The task lands on your board assigned to you; an agent only puts it in its own queue if you ask for that (create_task's assignee set to claude, or @claude in a quick capture).

2. How do I import a list I already have?

Try: “Put all these tasks into my Done board, grouped by project: [paste your list].” Paste anything — a notes file, an email, a spreadsheet column. The assistant reads your existing projects with list_projects, creates any that are missing with create_project, and files each line with create_task under the right project_id. It's an ordinary sequence of tool calls, not a special importer, so the agent can ask you about a line it can't place before it files anything.

3. How do I break a big task into smaller steps?

Try: “Help me break down the biggest task on my Done board into smaller steps.” The assistant looks over your open work with list_tasks (estimates and notes are what it judges “biggest” by), proposes a set of steps, and writes them in one call with decompose_task, which creates every step as a subtask under the original. The parent stays put, so nothing about your board's priorities moves until you decide it should.

4. How do I get the agent to actually work my list?

Try: “Let’s work on my Done board, highest priority first.” This is the prompt that turns a connected to-do list into delegated work. The assistant calls next_task, which atomically picks and claims the single highest-priority task that's ready — skipping anything blocked, paused, already waiting on your approval, or held by another agent — does the work, and reports back. On a task you've marked agent-ok it can close it with complete_task and a summary of what it delivered. On every other task — the default — complete_task is refused and it has to call propose_completion instead, which posts its summary and parks the task at pending approval. That's what shows up in Needs you: you approve it to done, or reject it back with feedback.

5. How do I make it run without me asking every time?

Try: “Every weekday at 8am, look at my Done board and work on the outstanding tasks.” Done doesn't run an agent on its own servers — the schedule lives in your assistant: a Routine in Claude, a Scheduled task in ChatGPT, or cron with a token for anything else. Each run starts by checking list_triggers for anything you asked for directly (Run now, or @claude in a comment), then works the queue with next_task, and stops at the same approval gate as prompt four. The guide to putting an agent on a schedule covers the runner prompt and the caps you can set on how much one run does.

What does the agent do when it can't finish something?

It hands it back rather than guessing. ask_decision posts a question to you with the agent's own recommendation, parks the task as needing your input, and the agent moves on to other work; it reads your answer on its next run. release_task drops its claim so the task goes back in the queue for a later try. Both surface in the app, so the morning after a scheduled run you see what shipped, what waits on your approval, and what waits on an answer only you can give.