Needs you: your agents' questions and approvals

Needs you is where your agents bring you the things they cannot decide on their own. The three oldest items sit at the top of the Agents Overview, with "See all" for the rest, and Needs you in the Agents panel lists every one, oldest first. Its count is the same number the Agents entry in the left rail shows.

Each item is one row: the agent, a tag for what kind of ask it is (Approve, Answer or Review), how long it has waited, the ask itself and a line of context. The button on the right acts on it straight away: Approve an approval, Reply to a question with a short answer box, Mark done a request for something only you can do, or Choose between two decisions that disagree. Once you act, the row goes and every count drops. When nothing is left, it says "You're all caught up."

For approvals there is more on the item than fits the row. Review on the Needs you page, or Open on the Overview's short list, unfolds the full card before you decide, including the items only you can do. Two things make it easy to act on: who does the work, and what led to it.

Who does the work

Every approval says whether the agent will do the work or whether it is yours to do.

• "Agent will do this" means the work runs the moment you approve. Approving is your go-ahead.
• "You do this" means it is something only a person can do, like connecting an account or confirming with someone. Approving marks it done because you did it, not the agent.

Every row shows how old it is, and anything still waiting after a week is tagged Stale, so a request from this morning does not look the same as one from last week.

Your three choices on an approval

• Approve. The agent runs the task, or the human task is marked done.
• Reject. The item is dropped. You can add a short reason so you remember why later.
• Request changes. Only for agent work. You leave a note saying what to change, and the agent runs again with your note in hand, instead of you rejecting a nearly-right result.

What led to this

Each card carries the context behind it: where it came from, a short excerpt of the step before it, and a link to open the full thread. Open the thread when you want the whole story before deciding.

If an item was created by a routine and a newer run of that routine has since happened, approving the old one does nothing. GPTree will not re-run stale work; it tells you the item was superseded.