title	First task
description	Plan, build and review one change: describe it, answer the questions, approve the Plan, then review and approve the result.
mdx	format mdx

import { Steps, Step, Tabs, Tab, Tip, Note, Screenshot, NextSteps } from "@site/src/components/kit";

Pick a small change in a repository you know. You describe it, answer a few questions and approve twice; your agent does the rest.

Plan, build and review {#steps}

Choose **+ › Task**, then **Feature** or **Bugfix**. Describe the change, attach files if needed, then choose your agent and where it runs. They come one at a time in **Your turn**. Each answer reaches your agent at once. Read it, then choose **Approve and build here**. To ask for a change, mark the Plan and write quick feedback. **Choose where to build** picks another agent or workspace. Follow Implement and Verify, or do something else. Anything only you can do reaches you as one question. On Verify, **Review the code**, see what changed and how **Done means** ran, then approve. **Pull request** on Verify opens GitHub's page to create it. You merge it there.

To reuse a plan you already have, choose Existing plan, paste or attach it, then Start from this plan (details). + › Chat starts a conversation without steps (Chat).

Where your agent runs

New task and Start the build both ask.

Recommended. Pick the agent, model, effort, project and branch. Questions and requests to your agent happen in the App. Ask your agent in your terminal, IDE or agent app in your own words, or copy the prepared prompt; attached files are listed by path. The agent asks in its own chat. You can still approve in the App, or [work only in your agent](work-in-your-agent.mdx).

Good to know {#good-to-know}

• Say QualityLayer or QL when you ask your agent to start work. Follow-up answers continue the task; no slash command is needed.
• Select a passage, write in the quick feedback field and press Enter to send it. Shift+Enter starts a new line.
• You can stop or pause the run. Your agent's own permission and sign-in prompts still need you.
• To continue elsewhere, stop the run and ask: "Using QL, continue task retry-webhooks." The Plan and evidence stay with the task.
• With several approved Plans waiting, "Implement this using QualityLayer" lists their commands.

<NextSteps items={[ { title: "The App", href: "/docs/app", text: "Home, the task page and its sidebar." }, { title: "Plan", href: "/docs/steps/plan", text: "What your agent asks and what you approve." }, { title: "Supported agents", href: "/docs/agents/native", text: "Models, permissions and reviewers per agent." }, ]} />