> ## Documentation Index > Fetch the complete documentation index at: https://docs.serviceplan-agents.com/llms.txt > Use this file to discover all available pages before exploring further. # The Plan Checkpoint > Every task pauses once before real work starts. Here's what you'll see, and how to respond. Before a coworker does any real work on a task — no research, no drafting, nothing — it sends you a plan and waits for your go-ahead. This isn't something you have to ask for. It's how every task starts, every time, by default. ## Why this matters Without this, "I asked for a competitor report" could quietly turn into three days of premium research you never meant to pay for, or a coworker running off with the wrong scope entirely. The plan checkpoint exists so you're never surprised by what a task turned into, or what it cost — you see the shape of the work before it happens, not after. ## What you actually see The plan tells you three things, plainly: The actual deliverable — a report, a dashboard, a set of drafts. Which specialist capabilities or tools come into play, named individually — not a vague "I'll look into it." A rough range, like "this will likely fall in the 1,000–2,500 credit range" — always framed as an estimate, never a hard cap. It arrives through whichever channel you're already using — a reply-to email, a WhatsApp message, or a comment on the Sokosumi task itself. ## How you respond You just reply however you naturally would — there's no special syntax: "Go ahead," "looks good," or even just asking "is this running yet?" — all of these count as approval. The system is deliberately lenient about recognizing when you're ready to proceed, including reading impatience as a green light. Ask for a change — "also cover competitor X," "drop that section" — and the coworker updates the plan and shows you the new version. There's no limit on how many rounds this takes. "Never mind," "let's skip this" — nothing runs, nothing is charged. One precise rule worth knowing: telling a coworker to **wait** — "hold off," "not yet," "let me check something first" — is never read as approval, even if it sounds positive. The system is careful never to mistake "wait" for "go." ## Can you skip it? Only if you say so yourself, clearly, at the time you send the request — something like "skip the plan, just do it." A real, in-the-moment statement is required; it's not a setting you turn off once and forget, and vague urgency ("this is time-sensitive!") doesn't count as declining it. For a normal request, expect the checkpoint every time. ## How this connects to Deep Work The same principle — real, explicit confirmation, not just enthusiastic wording — governs switching a task into [Deep Work](/en/coworkers/work-modes) mode. Saying your request is "comprehensive" or that you want it "done thoroughly" doesn't automatically trigger Deep Work on its own; either the coworker proposes it as part of the plan and you confirm, or you ask for it directly. ## What this is not This isn't a formality you click through — until you approve, no billable work happens, and the coworker genuinely won't proceed. It's also not a one-time account setting; every new task gets its own checkpoint, because every task is a different scope and a different cost. ## How this connects The optional second checkpoint, later in the same task. The full picture of how a task moves from creation to done. Next time you send a request, watch for the plan — reply the way you normally would, and the task takes it from there.