> ## Documentation Index > Fetch the complete documentation index at: https://docs.poly.ai/llms.txt > Use this file to discover all available pages before exploring further. # A full session, step by step > One build from first prompt to merged branch, with screenshots from a real session. These are the steps you'll repeat for most build work. The screenshots are from a real session, so what you see here is what you'll see in Studio. The example below adds a knowledge base topic, but the shape is the same whether you're writing a flow, wiring an integration, or changing how the agent speaks: **describe it, answer any questions, approve the plan, review what changed, then merge when you're happy.** The chat panel opens with a prompt input. Choose **See prompt suggestions** for starters grouped by intent: **Build**, **Test**, **Ask**, and **Analyze**. Pick a suggestion, fill in the blank where there is one, or ignore them and type your own request. Wren home screen with the composer and prompt suggestions grouped into Build, Test, Ask, and Analyze The more grounded your request, the fewer questions come back. *"Add a topic for our office opening hours"* works, but *"Add a topic for our office opening hours: 9am to 6pm Monday to Friday, closed weekends and UK public holidays"* gets you there in one turn. See the [prompting guide](/wren/prompting) for more. Rather than guess at business rules it can't infer, Wren asks. Questions come in a set you step through: pick an option, choose **Something else** to type your own answer, or **Skip** to let Wren decide. Wren asking a clarifying question with numbered options, a Something else option, a Skip button, and a 1 of 3 pager This is the step that stops Wren inventing your policies. If it doesn't know your refund window or your opening hours, it asks rather than picking something plausible. Once it has enough context, Wren presents a plan card summarizing what it intends to change, with **Approve** and **Revise** buttons. **Nothing is written until you approve.** Expand the card to read the full breakdown. Plan card titled Add Callback Request Flow with a summary, a what's changing section, and Revise and Approve buttons **Revise** opens a feedback box. Say what should be different and **Send** to re-plan, or leave it blank to reject outright. Plan revision feedback box reading What should be different, with Cancel and Send buttons For small, unambiguous changes Wren may skip the plan and just do the work. The bigger or more structural the request, the more likely you'll see a plan first. You can always force one by asking: *"Plan this before making any changes."* The first time Wren makes a change it creates a branch automatically, named after the request and prefixed `wren/`. `Main` stays untouched, and the branch chip replaces `Main` in the top-left for the rest of the session. Each step it takes is listed as it goes, with a running count of changes made. Transcript showing a branch named wren slash add-office-hours-topic being created, followed by four steps and a four changes made summary Expand the summary to see each step, and open any changed resource to check it by hand. Everything Wren wrote is an ordinary edit on your branch, so you can adjust or undo it exactly as you would your own work. Merging is yours to do. Wren never does it for you. The **Merge** button in the top-right opens a dialog where you can add an optional message, then ship the branch back to `Main`. From there, your changes flow through the normal [Sandbox → Live](/environments-and-versions/introduction) pipeline. Merge to the main branch dialog with an optional message field and a Merge button Not ready to merge? Leave the branch where it is. You can come back to the same chat later, keep iterating, and merge when it's right. ## What to try next Once you've been through this loop once, the rest of Wren is variations on it: Phrase requests so you get what you meant, in fewer turns. Every building block it can create and change. Ask questions about real calls, no branch needed.