Claude Code — AI-Assisted Dev Mastery

What you'll build

• Delegation, verification, and end-to-end AI-app workflows with MCP
• CLAUDE.md, checkpoints, and hook-based guardrails alongside skills
• Subagents, worktrees, and parallel specialist roles
• Planning mode, permissions, and runtime controls for risky change

Lessons

1. The CLI Agent Workflow

   Beyond the chat box. Master the Slash Command system, argument variables ($ARGUMENTS, $0), and shell injection (!prefix) that turn a CLI into a development partner.

2. Four tiers of persistent behaviour — prompt, memory, skill, hook

   Learn the four tiers of behavior. Master the Progressive Disclosure Architecture (Metadata -> Body -> Bundled Files) and build skills that scale without bloating your context.

3. Autonomous workflows — from idea to instrumented stack

   Treat Claude Code as the meta-skill of AI engineering: end-to-end prototyping, MCP-backed tooling, debugging retrieval and agent loops in your own code, and disciplined iteration — not one-shot magic prompts.

4. Memory layers, checkpoints, and hook-shaped guardrails

   Unify three operational layers: CLAUDE.md and rules as authored context, automatic checkpoints and rewind for session recovery, and event hooks (in settings) for enforcement — complementary to the skill/hook tiers in lesson 2.

5. Subagents — isolation, delegation, and parallel work

   Configure specialised subagents with their own tools, models, and permission modes; use worktree isolation and background execution; orchestrate review → implement → test pipelines without trashing your main context window.

6. Planning Mode, Permissions & Runtime

   The engine room. Master Claude Code's internal execution loop: from Planning Mode to Permission boundaries and background worktrees.

7. Plugins - architecture, components, and trust boundaries

   Claude Code plugins are not just packaged skills. Learn the full component model - skills, agents, hooks, MCP servers, LSP servers, and monitors - and design plugins with clear trust boundaries.

8. Plugin MCP servers and monitors

   Package live tools into Claude Code plugins with bundled MCP servers, and understand when monitors belong beside them. Learn startup behavior, configuration boundaries, and failure modes.

9. LSP servers and code intelligence

   LSP plugins give Claude Code live diagnostics, symbol navigation, and language awareness. Learn how to configure LSP servers correctly, what they do not include, and where teams misuse them.

10. Plugin scopes, debugging, and rollout

    Installing a Claude Code plugin changes runtime behavior. Learn installation scopes, common plugin failures, and how to debug and roll out plugin changes safely.

11. Build a Claude Code plugin workshop

    Build one Claude Code plugin end to end. Package commands, skills, agents, hooks, MCP, LSP, monitors, and tool boundaries into a coherent runtime bundle with concrete examples.

12. Repo-scale Claude Code - monorepos, context control, and parallel search

    Very large repos break naive chat workflows. Learn how to combine planning mode, grep breadth, LSP precision, subagents, memory files, and explicit context budgets so Claude Code stays useful inside monorepos.

13. Secure automation and plugin trust

    Claude Code plugins can execute shell commands, start background processes, and expose external tools. Learn least-privilege plugin design, shell-hook review, MCP credential boundaries, rollout approvals, and the failure modes that actually break trust.

14. Case study - repo config vs plugin-packaged runtime

    Compare two ways to share Claude Code behavior with a team: repo-local `.claude` configuration and a reusable plugin runtime. Learn the tradeoffs in portability, governance, debugging, and operational cost.