Learn 99% Of Claude Code in One Video

Summary

Tech With Tim presents a broad setup and feature guide for Claude Code, moving from installation and basic project use to advanced customization and workflow tools. The central advice is to configure the workspace and give Claude clear project context before relying on repeated prompts.

Setup and basic workflow

• Claude is available through the web, desktop, mobile, terminal, and IDE integrations. The presenter recommends the desktop app for most users and the mobile app for checking on agents or asking quick questions.
• Access to Claude Code depends on the account plan described in the video. The presenter says the Pro plan includes it, while higher plans offer more usage.
• Switch from chat to Code mode for software work, then open the project folder Claude should work in. Sessions are separate conversations within a project.
• The desktop interface includes controls for the terminal, changed files, an integrated browser for testing applications, and project files.

Models, effort, and permissions

Model choice involves balancing capability and usage. The presenter recommends the strongest general-purpose model for most work, with smaller models for simpler or repetitive tasks and to conserve usage. The model names and versions mentioned are specific to the recording and may change.

Effort level controls how much time and computation Claude spends reasoning and checking its work. Lower settings are faster and cheaper; higher settings are intended for difficult tasks where correctness matters. The presenter advises using the middle setting for everyday work and reserving the highest settings for demanding checks, such as security audits.

Auto permissions are recommended for normal use: Claude proceeds with routine work and asks before actions it considers risky. Other options include:

• Manual: Asks for approval for every tool call.
• Accept edits: Automatically accepts file changes.
• Plan: Prepares a proposal without changing files.
• Bypass permissions: Removes approval prompts and is discouraged unless the user understands the risks.

Planning and project context

For a new project, begin in Plan mode and describe the goal, preferred technologies, and conventions. Ask Claude to clarify requirements before it builds. Review the resulting plan—especially its proposed stack and file structure—before approving it.

The video demonstrates this process by planning and building a 3D scene, then previewing the result in the integrated browser.

CLAUDE.md (shown in the subtitles as claude.md) provides persistent project instructions, such as architecture, commands, and coding conventions. The presenter demonstrates creating one with /init; Claude can use it in later sessions. Instructions can also be placed in nested files for relevant parts of a project.

Claude can save memory when asked, including information in local, project-specific files. The presenter distinguishes local memory from information intentionally kept in the project repository.

Project-level settings.json can define permissions and behavior for a repository—for example, commands Claude may run, actions that require approval, and files it must not read or edit.

Add context with images, files, folders, or the @ reference syntax. The desktop interface also supports commenting on code and annotating parts of a browser preview.

Reusable tools and extensions

• Skills are reusable instruction sets, useful when a task repeatedly needs the same steps or output format. They can be created by asking Claude, invoked with a slash command, and stored globally or for a single project.
• Subagents handle delegated tasks in separate background contexts, with configurable instructions, tools, model, and effort. This can keep research, testing, reviews, or other work from consuming the main conversation’s context; multiple agents can run in parallel.
• MCP servers (Model Context Protocol) connect Claude to external tools and services. They can be managed through the interface or added by asking Claude to configure a connector.
• Hooks run commands in response to Claude Code events, rather than waiting for Claude to choose to run them. They can be configured globally or per project; the demonstration uses a hook to show a desktop notification when a session finishes.
• Plugins bundle related capabilities, potentially including skills, MCP configuration, and subagents.
• Mods are described as a newer, more extensive way to customize Claude Code’s interface or behavior, such as by adding panes, status displays, notifications, or hooks.

Context and working with changes

The context panel shows what is occupying the conversation’s context window. The presenter recommends monitoring it because long contexts can increase usage and may make performance less reliable.

Useful slash commands include:

• /context to inspect context usage.
• /clear to clear the current conversation context.
• /compact to summarize and reduce conversation history. Claude may also compact automatically.

The up-arrow key can retrieve earlier prompts.

Rewind can restore an earlier point in the conversation, undoing the conversation, code changes, or both. The presenter notes that a rewind can also be undone.

Multiple sessions can run at once. Git worktrees give sessions separate copies of the code to reduce interference, though merging their changes can still produce conflicts.

Code review can examine changes in a Git-connected project without making edits.

Routines are scheduled tasks that can run locally or in the cloud, subject to the setup described. Examples include PR reviews, release notes, health checks, briefings, and email triage.

Guides and demonstrations covered

1. Installing Claude Code and choosing where to use Claude.
2. Opening a project and navigating the desktop Code interface.
3. Selecting models, effort levels, and permission modes.
4. Planning a project before implementation.
5. Creating project instructions, memory, and settings.
6. Supplying context through files, references, and visual annotations.
7. Creating and using skills, subagents, MCP connections, hooks, plugins, and mods.
8. Managing context, rewinding changes, running parallel sessions, reviewing code, and scheduling routines.

Main speaker and sources

• Main speaker: Tim, from the Tech With Tim channel.
• Other source: Anthropic is identified as the video sponsor; Claude and Claude Code are the products being demonstrated.