What's new in the course

November 2026 · full content review

New lessons

• Auto mode, sandboxing and checkpoints (Module 4): the classifier, autoMode.environment, /auto-mode-setup, ask-rule checkpoints and the Bash sandbox
• Parallel sessions: worktrees and agent view (Module 7): --worktree, claude agents, --bg, /fork, /background, cross-session messaging
• Long-running work: /goal, /loop and scheduled tasks (Module 8)

Updated for current Claude Code behaviour

• Auto mode is now the default permission mode for interactive sessions, and Module 1 explains every mode, including dontAsk
• Checkpoints and /rewind: restore code or conversation, and summarise part of the history
• Custom commands are now skills: the .claude/skills/ layout, 0-based $0 arguments, disable-model-invocation, and allowed-tools pre-approving rather than restricting
• Memory: .claude/rules/ with paths, auto memory, AGENTS.md support, and /doctor prompt-audit
• Permission rule syntax Bash(cmd *), parameter rules like Agent(model:opus), and protected paths
• Hooks: over 30 events, the if filter, http/prompt/mcp_tool hook types, and why exit 1 doesn't block
• Models: Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.1 aliases, /effort levels, ultrathink, opusplan and /fast
• Subagents run in the background by default; /agents is no longer a wizard; new /subtask, isolation: worktree and /batch
• Built-in review commands: /code-review, /simplify, /security-review and ultrareview
• Headless: --bare, --json-schema, dontAsk and --max-budget-usd
• GitHub Actions @v1 workflows, the code-review plugin and the managed Code Review service
• Plugins: claude plugin validate, claude plugin eval, namespacing, userConfig and bin/
• Claude in Chrome (--chrome), claude mcp login, MCP tool search, and Windows install without Git Bash