Course Description

Claude Code 2.0: Long-Running Agents, Automation & Cloud Execution

With the release of Claude Sonnet 4.5 and Claude Code 2.0, AI coding agents have crossed a major threshold. Claude is no longer just a terminal assistant — it can now run long-running tasks, act as a general autonomous agent, and even work entirely in the cloud via GitHub, without you being present.

In this lesson, you’ll learn how to leverage Claude Code’s new capabilities to build automated agents, GitHub-hosted workflows, and cloud-based coding tasks that run in the background for hours. You’ll also learn the critical rules required to avoid hallucinations, permission failures, and broken automations when working with MCPs and GitHub Actions.

What You’ll Learn

• What makes Claude Sonnet 4.5 the strongest coding model so far

• How Claude Code 2.0 handles long-running, multi-hour tasks

• Why Claude is evolving from a coding agent into a general-purpose agent

• How to run Claude Code autonomously using GitHub Actions

• How to receive automated notifications (Slack, external services)

• How to run background coding tasks from your phone

Use Case 1 — Claude Code as a General Automation Agent

You’ll build a fully automated AI model release monitor that:

• Runs automatically twice a day

• Monitors multiple AI provider websites

• Sends Slack notifications when new updates appear

• Can also be triggered manually via GitHub Actions

You’ll learn:

• How to host Claude Code as a GitHub App

• How to initialize Claude inside a repo using a guided slash command

• How to define GitHub workflows using plain-English prompts

• Why most failures come from missing tool permissions

• How to safely add MCPs for cross-platform automation

Critical Rules for Reliable GitHub Automations

You’ll learn two essential rules that prevent “hallucination hell”:

• Rule 1: Always auto-approve tools

  • GitHub workflows cannot wait for manual approvals

  • Missing approvals silently break automations

• Rule 2: Always test MCP servers

  • Verify MCP functionality before relying on it

  • Avoid incorrect assumptions about third-party APIs (e.g. Slack)

Use Case 2 — Cloud-Based Coding Agents (Codex-Style)

Inspired by OpenAI Codex, you’ll learn how to:

• Open GitHub issues as tasks for Claude

• Tag Claude to automatically start working

• Let Claude implement features end-to-end in the background

• Handle full coding workflows without your local machine

You’ll see a real example where Claude:

• Implements authentication in a GitHub-hosted app

• Uses Supabase MCP to automate setup

• Configures backend services without manual API handling

• Commits changes and confirms completion autonomously

Working with MCPs in Cloud Agents

You’ll learn:

• How MCPs extend Claude’s capabilities across platforms

• Why missing SDK documentation breaks cloud agents

• How to fix stalled agents by providing official SDK docs

• How to identify broken or read-only MCPs

• How to safely replace MCPs with full read-write access

New Claude Code 2.0 Features Covered

• Rewind Mode

  • Restore previous conversation states

  • Optionally revert code changes as well

• Improved IDE Extension

  • Cleaner UI than terminal mode

  • Plan mode, slash command menus, file context visibility

  • Better overall usability for long sessions

Who This Lesson Is For

• Developers using Claude Code who want true automation

• Builders creating background agents and workflows

• Anyone deploying AI agents on GitHub

• Developers tired of babysitting AI coding sessions

• People who want cloud-based, long-running AI execution