An approval lets you review the proposed action, account, destination and content before it happens. Approve authorizes the revision shown; it does not mean the work has already completed.
What the card shows
The first screen answers what will happen, where, from which account and to how many people or objects: a title with the concrete outcome ("Review email send", "Review HubSpot contact create", "Review Skill plan"), a short block of facts (recipient, account, destination, count, schedule), the content itself, and one row of controls. If the message could not carry everything, View full details in the ··· menu shows the whole reviewed text.
What the buttons do
A card has one primary button that names the outcome — Approve & send, Approve & create, Approve & write, Approve plan — and at most one secondary button. Everything else sits in the ··· menu.
- Approve & …: authorize this reviewed revision. The card shows that work is starting, then the actual result: completed, partial, failed or unknown. A result can include links to what was created.
- Edit (secondary, on a draft that can be edited in place): change any field the action will use — for an email the recipients, subject and full body. Save changes produces a revised draft awaiting approval; it does not execute the action. Closing the editor without saving leaves the draft alone.
- Request changes (secondary on a plan, in the ··· menu on an editable draft): tell Nisa what to revise. The old draft becomes blocked, your words stay on the card, and the revised draft returns for a new review; it is not automatically approved. On a Skill's plan the work resumes with your words and proposes again; nothing from the old proposal runs.
- Don't send / Cancel (in the ··· menu): stop unstarted work without having to explain why. It does not undo something already executed. It is never red: stopping prevents the risk.
- Reject is red only where the click itself changes a record (an outreach sequence marked rejected).
Skill and outreach reviews have different capabilities. Only rely on the controls actually shown. A Skill plan offers Approve plan, Request changes where its saved work can resume with feedback, and Reject and View full plan in the ··· menu; it never offers an Edit that could not bind the saved action. A grouped outreach review lets the owner select one item and open its complete saved draft with Edit selected, or use More → Request changes for feedback and a revised pending draft.
A decision stays visible as a compact receipt: the state (Completed, Failed, Cancelled, Changes requested, Expired, Replaced by a newer draft), the reviewed content, who decided and when, and what to do next — a failed action says a retry is a fresh review and offers no retry button. A click on a card that was already decided elsewhere answers with that receipt and removes the stale controls. A stale card or old editor cannot approve a newer revision.
What an approval covers
One approval covers the action described on that card. A finite artifact task can group a named set of creations, such as a page and a spreadsheet. It does not authorize future recurring work or unrelated sending, sharing or spending.
Do not submit the same action again just because a receipt is delayed. A partial result retains what succeeded. An unknown outcome needs checking before any retry; it is not a reason to create another copy automatically.
Allow future … — a standing permission, confirmed on its own screen
Action cards offer Always allow… in the ··· menu. Selecting it never enables anything: it opens a screen that names the exact scope (what, where, which account, who it applies to, how long, how to undo, what stays outside it) and says that Allow & run also runs the pending action once. Keep asking me closes it and changes nothing. A workspace-wide rule can only be set by a workspace owner or admin; a personal create permission covers only you.
Send cards can offer future permission too. A held cross-room reply is scoped to its source set, owner and destination channel; a team-skill import to its source; and a purchase to its merchant, exact amount, cap and paying account. Those three grants cover only the card owner. A different scope asks again. Login handoffs, bank challenges and permission-confirmation cards still need a human action because their click attests to something only the person can do.
You can also ask Nisa in chat to allow supported creation in a specific destination. Nisa presents a separate confirmation card. Nothing is enabled until you click Allow.
The permission covers only you in that workspace, one connected account, one exact parent destination and the artifact types named on the card. It does not include child folders, teammates, guests, scheduled Skills, editing existing items, sending, sharing, deleting or spending. It can expire or remain until revoked. Ask Nisa to list your create permissions or revoke one; there is no separate settings screen for these new grants.
Supported types are Notion pages, Google Docs, Google Sheets and Drive files. Google Doc creation permissions currently cover My Drive root only, using the same connected account. They cannot promise creation in a selected folder. Other supported types use their named parent destination.
A finite task skips the approval card only when every item is covered. The permission is checked again before starting, and before each child of a task. Revoking it prevents further work under that grant; it does not undo completed work.
These grants are not a permission catalogue for every tool. Other tools retain their existing workspace policies, including reversible actions that may already run without a card. Revoking a create grant does not change those separate policies.
When Nisa builds a Skill
A Skill should ask once for the decision you actually want: review of its report, or review before a consequential action. Those are different gates. After a decision, Nisa carries on with the saved work rather than restarting the original request and repeating completed actions. A personal create permission does not grant a recurring Skill autonomy or bypass its outbound review.
