Claude 2.0 완전 정복! 왕초보를 위한 클로드 코드 데스크탑 새 기능 13가지 (15분 정리)

Summary

The video gives a beginner-oriented tour of 13 features and interface changes in the Claude Code desktop app. Its central point is that the app is evolving from a place to ask questions into a workspace for delegating, tracking, and reviewing tasks.

Features and Guidance

1. Sessions and permissions — Start a session for a particular task and choose the project folder it can access. The presenter recommends keeping permission requests on manual at first rather than enabling “Ignore Permissions,” so file changes require approval.

2. To-do and status cards — Claude can display a task list and mark items complete. Permission cards show when Claude needs approval; if work appears to have stopped, check for a pending card.

3. Context card and /compact — The context card shows how much conversation information is in use. /compact summarizes earlier conversation to free up context. The presenter often uses it when context is roughly 30–40% full.

4. Sections — Group sessions by topic, such as by video series, to make older work easier to find.

5. Sub-sessions — A main session can delegate separate tasks to child sessions and collect their results. Unlike sub-agents, which Claude uses temporarily, sub-sessions can be opened and addressed directly. The presenter recommends using them to split up substantial or unrelated work.

6. Side chat — Open a separate, lightweight chat based on the current session. It’s useful for questions or revisions that shouldn’t be added to the main conversation, especially when Claude is busy with a longer task.

7. Split view — Show two sessions side by side to compare work or monitor separate tasks. The presenter uses it to compare a script with an editing session and check whether Claude made a requested change.

8. Artifacts — Create and review dashboards, slides, documents, and designs alongside the conversation. Users can comment on specific parts to request edits, then check the result and its sharing settings before showing it to others.

9. Slash commands — The tutorial highlights four:

   • Output styles set response tone and length. The presenter demonstrates a concise style using simple language.
   • /doctor checks configuration and instruction files for potential issues.
   • /skill-doctor reports on skills and their context usage, including unused skills.
   • /advisor lets a lighter model, such as Sonnet, consult a stronger model, such as Opus, for important decisions.

10. Modes and hooks — Modes add custom rules to the app’s behavior and are distinct from permission modes. One example is a hook that warns about text resembling a secret key before it is sent. The presenter also uses a hook to prevent two video renders from running at once.

11. Browser preview — View a webpage created by Claude in the app and check updates after requesting changes, without leaving the conversation.

12. Terminal — Open a terminal to run commands. Claude may ask the user to run a command directly; the presenter recommends asking Claude to explain unfamiliar commands before executing them.

13. Scheduled tasks — Set recurring work, such as a weekly summary, to run at a chosen time. The presenter advises reading the scheduling permission prompt and testing the task before relying on its scheduled run.

Additional Capabilities and Advice

The app also shows the active project environment, can connect to other services, and can upload files created or changed during a session.

The video’s closing advice is to start with one frequently repeated task and try the feature that best fits it, rather than learning everything at once.

Main speaker/source: Sam Hottman, presenting a hands-on walkthrough of the Claude Code desktop app.