Guides / Build a Review Screen
Build a Review Screen
Put your review queue on a wall, in Slack or in your admin, driven entirely by each card's actions.
The review queue is the agent's inbox for you. Putting it somewhere you already look, a TV in the office, a Slack channel, a tab in your own admin, is the single biggest thing you can do to keep work shipping.
The rule that makes it easy
Every card carries two fields:
actions: the exact verbs the API accepts for this card right now.primaryLabel: what the dashboard's main button would say, for example "Send Reply" or "Approve + Publish".
Render one primary button with primaryLabel and a secondary control per remaining verb. Never branch on sourceType or kind; the rules that decide what a card accepts live in one place on the server and change without you.
Verb in actions |
Call |
|---|---|
approve, answer, choose |
POST /review/{id}/answer with action set to the verb |
send |
POST /review/{id}/send |
request_changes |
POST /review/{id}/request-changes |
snooze |
POST /review/{id}/snooze |
dismiss |
POST /review/{id}/dismiss |
Polling
GET /review returns pending and needs-revision cards, newest first within priority band. Every 10 to 30 seconds is plenty and far inside the read limit. Diff by id and updatedAt to animate changes.
Handling a 409
Two people, or a person and the trust ladder, can act on the same card. A verb the card no longer accepts is a 409 whose body carries the current actions. Re-render from that and move on; nothing was double-applied.
Agents: this page is also markdown at /docs/build-a-review-screen.md, answers Accept: text/markdown, and is listed in /docs/llms.txt.
Frequently Asked Questions
Yes. `GET /review/{id}` returns the card's payload with long text capped at 8 KB per field, which is enough for any draft the agent writes.