> ## Documentation Index > Fetch the complete documentation index at: https://skyvern.com/docs/llms.txt > Use this file to discover all available pages before exploring further. # Your first agent with Copilot > Describe a browser task in plain words on the Skyvern Home page, let Copilot build the agent in Studio, accept its changes, run it and read the results. You describe a task in plain words. Copilot turns it into an agent, tests it in a live browser and hands it back for you to accept. Then you run it and read the results. ## Build and run your first agent Create an account at [app.skyvern.com](https://app.skyvern.com). A new account answers a few short questions first (you can click **Skip for now**), then lands on **Home**. If you described a task in those questions, it is already in the prompt box. In the box under **What task would you like to accomplish?**, write the task the way you would brief a colleague. Name the site, say what to do there, and say what you want back. ```text theme={null} Go to news.ycombinator.com, open the top 5 posts on the front page, and return each post's title, URL and point count. ``` Click the arrow to send. * **Try an example** below the box fills in a ready-made prompt you can edit. * The **+** menu attaches files to the prompt (**Upload document**) or starts a recording of you doing the task yourself (**Record task**). * The microphone button lets you dictate the prompt. Home with the Hacker News task typed in the What task would you like to accomplish box and the Try an example cards below Skyvern creates a new agent and opens it in **Studio**, with your prompt sent to **Copilot**. You don't need to do anything yet. Copilot names the agent, plans the steps, opens the site in the **Browser** pane and adds blocks one by one. It runs each block to test it, so you see the browser move while it works. If Copilot only answers in the chat without proposing blocks, ask it to turn the work into an agent you can run again. Copilot stops to ask when it needs you: * **A question.** Pick an answer or type your own in the message box. * **A login.** The **Copilot needs to sign in** card lets you pick a saved login under **Saved logins** or add one. See [Credentials and 2FA](/docs/credentials). Studio with Copilot showing its plan and working status on the left and the Browser pane on Hacker News on the right When Copilot finishes, it shows a **Proposed changes** card listing the blocks it added, changed or removed. A pill tells you whether the blocks were **Tested**, **Untested** or **Test failed**. * Click **Accept** to apply the changes. They are saved to the agent. * Click **Review** to compare the proposed agent with the previous version before you decide. * Click **Reject** to discard them. The agent goes back to the previous version. Nothing is saved until you accept. To stop reviewing every change in this chat, open the arrow next to **Accept** and choose **Always accept**. The message area then shows **Auto-accepting** with a **Turn off** link. Copilot reply with the extracted posts and a Proposed changes card marked Tested, with Accept, its Always accept option, Review and Reject **Test end-to-end** (in the card's **⋯** menu, or in the card itself) runs every block together on the real site. It can submit forms, place orders or send messages for real. Click **Run** in the Studio top bar. The run form opens. Fill in any inputs the agent needs, then click **Run agent**. The default settings work for a first run. [Runs and schedules](/docs/cloud/runs#run-an-agent) explains every setting. Studio switches to the run. The **Browser** pane shows the browser live, and the **Run** pane's **Timeline** shows each block as it runs. When the run ends, open **Outputs** in the Run pane. You'll find the data the agent returned (**Extracted information** and **Run outputs**), any **Errors**, and links to **Downloaded files**. A completed run with the Run pane on Outputs showing the extracted Hacker News posts You now have a saved agent. Find it any time under **Agents** → **All Agents**, run it again, or [put it on a schedule](/docs/cloud/runs#schedule-an-agent). ## Change the agent by chatting Keep talking to Copilot in the same pane. Ask for changes the way you asked for the agent: ```text theme={null} Also return the number of comments for each post, and skip posts that are job ads. ``` Each request ends in another **Proposed changes** card to accept or reject. **New chat** at the top of the Copilot pane starts a fresh conversation, and **History** reopens earlier ones. If a run fails, the Run pane shows what went wrong with a **Fix with Copilot** button. Click it to hand the failure to Copilot, then review its fix like any other change. ## Other ways to start | Start from | How | What you get | | - | - | - | | A template | On Home, scroll to **Start from a template** and click a card | A copy of the template named "*template name* (copy)", open in Studio. Copilot posts what it does and which inputs it needs. | | Your own template | **Agents** → **All Agents** → **Create** → **From Template** | A copy of an agent your team saved with **Save as Template** | | Nothing | **Start with a blank canvas** under the Home prompt, or **New Agent** in the sidebar | An empty agent in Studio. Ask Copilot or add blocks by hand. | | A file | **Import** on the **All Agents** page | An agent from a YAML, JSON or ZIP export, or generated from a PDF procedure | [Build in Studio](/docs/cloud/studio) covers editing blocks by hand, inputs, versions and organizing agents. ## Tips for a good first prompt * **Name the site.** A full URL saves Copilot a search. * **Say what "done" looks like.** "Return the title, URL and point count" works better than "look at the posts". * **Say what varies between runs.** "The search term changes each time" tells Copilot to make it an input. * **Mention logins up front.** If the site needs an account, say so, so Copilot asks for a saved login early. ## Next * [Build in Studio](/docs/cloud/studio): edit blocks, inputs and versions yourself. * [Runs and schedules](/docs/cloud/runs): run settings, live control, run history and schedules. * [Credentials and 2FA](/docs/credentials): save logins and 2FA codes your agents can use. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.