Carrier Log in
← All tutorials

BasicsChatSurveyScripted agents

Build a simple study with an assistant

Build Harbour pilot, a group-chat study with two conditions, in six requests to a connected assistant. After each request, check the draft in the builder. You do the last step: the completion redirect and Deploy.

6 steps 6 min read

  1. Describe the study in one message

    Ask for a draft called Harbour pilot with two conditions, Control and Prompted, assigned at random. In each condition, one chamber for three people has:

    • Briefing (instruction).
    • Discussion (a text chat of ten minutes).
    • After the chat (survey).

    The assistant checks the change without a save, then saves it. It gives you the builder link.

    In the builder, each chamber line shows the three segments. The Max Duration of Discussion is 600 seconds.

    Two chamber lines, Control and Prompted. Each chamber has Briefing, Discussion and After the chat.

  2. Give it the briefing text

    Give the assistant your briefing text. Ask it to put the text in the two Briefing pages as Markdown.

    The assistant makes one patch with two operations. Thus Carrier saves the two changes or neither.

    Click Briefing, then click Edit content to see the text. In History, the newest change by the assistant is "Changed 2 settings", with the part chamberlines.

    The briefing text in Edit content. History shows "Changed 2 settings".

  3. Give it the survey questions

    Ask for three questions in the two After the chat surveys:

    • Two required ratings from 1 to 7.
    • One optional comment box.

    The assistant reads the Survey.js reference, writes the questions and saves.

    Click After the chat. The Survey Builder shows the three questions with their types.

    The Survey Builder of After the chat: two Rating questions and one Long Text question.

  4. Add the scripted participant to Prompted

    Ask for a scripted agent called Harbour host, only in the Prompted chamber. After the fourth chat message, it says "What would change your mind?" one time.

    The assistant creates the agent, puts it in the cast and adds one message-count trigger. A scripted agent needs no system prompt. Thus the assistant asks you nothing here.

    Click Harbour host, then click Open Script Builder. The trigger fires after 4 messages, max 1.

    Harbour host in the Prompted cast. Its one trigger: message-count, after 4 messages, max 1.

  5. Check the draft

    Ask the assistant to check the draft and explain each error and warning. It does the same checks as Deploy.

    For this study, expect no error and one warning: "Completion redirect is not set". An error stops a save or Deploy. A warning does not.

    Deploy, then Update: the same warning about the completion redirect.

  6. Finish in the builder

    The assistant cannot do these steps. Do them in the builder:

    • Click the study name. Experiment Settings opens.
    • Type your completion URL in Redirect URL.
    • Read the briefing, the three questions and the sentence of the agent again.
    • Click Deploy, then click Update.

    After you select Go live on the dashboard, the assistant can read the study but cannot change it.

    The completion URL in Redirect URL, then Deploy and Update with no warning.