Your agent reads our introduction written for assistants, summarises Mandarine, asks you its questions one at a time, then tells you frankly whether it fits your team — and prepares your first workspace.
It reads our introduction for assistants: https://www.mandarine.tech/en/for-your-agent.md
07Templates
"What can I ask my agent for?" A template answers: a post-mortem, a story to validate, a client request to review before sending. Each is a polished HTML skeleton the agent fills in, with the kit's buttons already in place. And nothing is imposed: a blank HTML page is still a card.
In Mandarine, a template is an HTML skeleton with named slots and the kit's components already in place. The agent fills it and publishes the card; Mandarine warns it if a slot was left empty. Styles come from a single stylesheet, versioned with the kit. The team's theme (colours, font, logo) applies to every card and stays frozen at its publication.
Context, in three linesThe single node restarted twice in October: 40 minutes without search in total. Three nodes cost a third more per month.
The agent fills the “decision” template skeleton, then publishes the card.
Named slots, in brackets, each with its instruction. The agent replaces, removes an empty section, adds a useful one. Whatever stays in brackets is flagged at publication.
Masthead, callout, timeline, stats, tabs, phone frame, before / after: one stylesheet, versioned. The agent's page stays short, and quality is tuned in one place.
Your colours, font and logo, written in the workspace's documentation. Every card inherits them on the day it is published, and keeps that look: a mockup validated in March stays March's.
If no template fits, the agent writes free HTML with the kit's stylesheet. The template is a starting point, not an obligation. The cards in the home page gallery show it: each has its own style.
The twelve
Each template says what the agent does and what you do in the page. Your agent knows the list, when to use each one and when not to: tell it the gesture, it chooses.
A closed question, two to eight options, with the agent's recommendation. "Should search move to three nodes?"
A vague request becomes a story. The agent writes what it understood and asks three to five questions before writing.
A user story ready to develop: role, need, reason, acceptance criteria, out of scope.
Two to four variants of a screen, desktop and mobile, what changes between them, and a vote.
A clickable flow of several screens, to be reviewed by the team or by an invited client.
A deal tracked CRM-style: stage, amount, contacts, history of exchanges, next action.
An incoming request, what the agent understood, the prepared reply and quote, the points of attention.
The analysis after an incident: impact, timeline, root cause, what went well and badly, proposed actions.
A gap found while running the test plan, with its screenshot: scenario, expected, observed, environment, proposed severity.
Monday's digest: three things to remember, then one block per theme, each item with its source and why it matters.
A project's starting documentation, in several pages: what it is about, vision, decision log, conventions, glossary.
Your team's identity, readable by all, and the source of the theme: colours with their contrast, typography, logo, tone.
Shown for real
Two templates filled by the agent, with fictional names. The kit's components are already in place: the vote, the checkboxes, "I'm done".
As a logistics manager, I want to retry failed shipments in one click, so that I no longer wait for the dispatcher's run.
The agent writes the team's design system in the workspace's documentation: colours, font, logo. Mandarine checks the contrasts and makes it the theme of every card. Try it with a fictional brand.
Frozen at publication: the card keeps the theme it had on the day the agent published it. A rebranding does not rewrite history.
The single node restarted twice in October: 40 minutes without search in total. Three nodes cost a third more per month.
The kit
They are what makes the agent's page readable by Mandarine: "14 answers out of 20", "an agent is waiting", "I'm done". Every answer is signed, saved as you go, and read back by the agent.
questionA free answer, short or long.
choiceOne option among several, with a word of explanation if wanted.
multichoiceSeveral options at once.
quick-replyAccept, edit or skip, finding by finding.
approvalYes or no, and why.
checklistShared between people and agents: every tick is signed and dated. The agent ticks what it has done.
voteThe whole team decides, one vote each. The agent counts at closing.
The button on every card. It tells the agent your answers are complete: it reads them, starts its work, and summarises it for you. On a card with no question, it becomes “Close the card”.
In its chat: "Write me a story for the accounting export", "Write yesterday's post-mortem". No need to name the template.
As soon as it is connected, it knows the list, when to use each one and when not to. If the request falls outside, it starts from a blank page.
The card lands in "To decide". You answer, "I'm done", and the agent reads your answers. "All the post-mortems" are one filter away.