Claude Code for Real-World Projects is a hands-on course in using Claude Code to deliver project features from issue to reviewed pull request, covering CLAUDE.md, Plan Mode, subagents, skills, MCP, hooks, permissions and CI. It suits software developers, tech leads and DevOps engineers who want an AI coding agent to work to their team's standards of quality and safety.
AI coding agents are no longer just tools that complete the next line of code. They behave more like a teammate who reads the whole project, plans, edits many files, runs commands and opens pull requests on their own. Many developers have started using them, but still work one prompt at a time, so the code comes out inconsistent, without tests to back it, with no control over permissions or spending, and hard to scale across a team.
This course teaches you to use Claude Code systematically, from setting up a project, designing context with CLAUDE.md, planning with Plan Mode and developing with tests in place, through to extending Claude Code with subagents, skills, MCP, hooks and plugins and connecting it to Git, GitHub and CI/CD. Every day builds on the same practice project, the NovaDesk Service Platform, until you can deliver a feature from issue to reviewed pull request quickly and with quality under control. (3 days, 6 hours per day, 18 hours in total, Intermediate level.)
Course Details
This is a hands-on course that uses Claude Code to deliver one real project from issue to reviewed pull request, with team governance through hooks, permissions, plugins and CI. It runs for 3 days, 6 hours per day (18 hours in total, 09:00-16:00), as lectures with labs and workshops built on one practice project, the NovaDesk Service Platform, which grows day by day. Intermediate level. The main examples use TypeScript, but you can do the workshops in the language you know best. Learners use their own Claude Pro, Max, Team or Enterprise plan or their own
API key; the institute does not provide accounts or subscriptions. It differs from Claude Code Roadmap, which covers Claude Code broadly over 5 days, AI Coding Agent for SDLC, which covers several AI coding agents across the development life cycle, and Build AI Multi-Agent with Claude Code, which focuses on directing advanced agent teams. It does not teach programming or Git from scratch, and does not go deep into building applications with the Claude API or Agent SDK or writing your own MCP server. Claude Code features change quickly, and the content follows the latest version on the training day.
Section 1: Agentic Coding and an Overview of Claude Code
How autocomplete, chat assistants and coding agents differ
The agent loop: read code, plan, use tools, check results and repeat
Where to use Claude Code: the terminal, VS Code, Desktop and the web
Choosing Opus 5.5, Sonnet 5.5 or Haiku 5.5 for the task, and the effect on speed and usage
Lab: explore the NovaDesk starter repository with your first questions to Claude Code
Section 2: Lab: Installation and Initial Setup
Install Claude Code and sign in with a Claude account or an API key
Settings levels: managed, user, project and local, and their order of precedence
Permission modes: Default, Accept Edits, Plan Mode and Auto Mode, and when to use each
Everyday commands: /init, /model, /config, /usage (or /cost) and /doctor
Lab: set up Claude Code in the terminal and in VS Code, ready for the project
Section 3: Context Engineering with CLAUDE.md
Writing a good CLAUDE.md: architecture, build and test commands, code standards and what not to do
CLAUDE.md at user, project and subfolder level, and using an existing AGENTS.md
Referencing files with @ and adding screenshots or supporting documents
Common mistakes: too much context, conflicting instructions and outdated information
Lab: create the NovaDesk CLAUDE.md with /init and tailor it to the project
Section 4: Managing the Context Window and Sessions
Check context with /context and use /compact and /clear at the right time
Resume sessions with /resume and keep separate sessions per task
Checkpoints and /rewind when the result is not what you expected
Writing clear prompts: goal, constraints and success criteria
Lab: rewind an edit that went the wrong way and give a clearer instruction
Section 5: Workshop Day 1: Plan Mode and the First Feature
Use Claude Code to explore and explain an unfamiliar codebase
The workflow: Explore, Plan, Implement and Verify
Review and revise the plan in Plan Mode before approving any code changes
Have Claude Code write and run tests to verify its own work
Workshop: build the "create and search tickets" feature for NovaDesk with all unit tests passing
Section 6: Lab: Development Workflow in a Real Project
Test-driven development with Claude Code: write the tests first, then let the AI make them pass
Debug from error logs, stack traces and unexpected behaviour
Refactor legacy code with technical debt safely, protected by tests
Review the diff every time before accepting a change
Lab: fix bugs and refactor an existing NovaDesk module with tests passing at every step
Section 7: Claude Code and the Git Workflow
Have Claude Code create branches, write commit messages and summarise changes
Create pull requests with high-quality descriptions
Git worktrees for running several sessions in parallel without collisions
Resolve merge conflicts with help from Claude Code
Lab: open the first pull request from the Day 1 feature branch
Section 8: Lab: Custom Slash Commands and Skills
Custom slash commands for repeated work, such as creating an endpoint or writing tests the team's way
Skills: bundling instructions, steps and reference files for Claude to use when relevant
The structure of SKILL.md and writing a description that makes the skill load at the right time
Share commands and skills with the team through the .claude/ folder in the repository
Lab: build a team skill for creating NovaDesk endpoints to standard
Section 9: Subagents and Splitting Complex Work
What a subagent is and how it keeps the main session's context clean
Build specialised subagents such as a code reviewer, test writer and documentation writer
Set the tools, model and effort for each subagent
Patterns for parallel work and combining the results
Lab: create code reviewer and test writer subagents for the project
Section 10: Workshop Day 2: MCP and the Assignment Feature
MCP concepts and local versus remote connections
Add and manage MCP servers at user and project level
Connect a database so Claude Code can read the schema and check sample data
Connect GitHub and up-to-date library documentation, with the security precautions involved
Workshop: build "assign and track ticket status" test-first, splitting API, frontend and test work across subagents, then open a fully described PR
Section 11: Lab: Hooks for Automatic Quality Control
Key hook events: PreToolUse, PostToolUse, UserPromptSubmit and Stop
Run formatters, linters and tests automatically after file edits
Block dangerous commands with exit code 2 or a JSON decision, and make a failing hook block
Check and test hooks through the /hooks menu before relying on them
Lab: a hook that formats files after edits and protects .env files
Section 12: Security, Permissions and Governance
Allow, ask and deny permission rules for commands and files through /permissions
Protecting secrets and .env files, and the risks of prompt injection
The sandbox that limits file and network access for shell commands on macOS, Linux and WSL2
Managed settings for organisations, and /security-review for a first vulnerability check of a branch
Section 13: Plugins and Sharing Team Standards
Plugins: bundling skills, subagents, hooks and MCP servers into one unit
Install plugins from a marketplace with /plugin and create a team marketplace
Share .claude/ and CLAUDE.md through Git so the whole team works to one standard
Lab: package the commands, skills and subagents you built into a team plugin
Section 14: Headless Mode and CI/CD Integration
Run non-interactively with claude -p and text, json or stream-json output
Claude Code GitHub Actions for reviewing pull requests and responding to @claude in issues
Automated code review with /code-review and setting the scope of review findings
Controlling permissions, allowed tools and secrets in the pipeline
Section 15: Costs and Team Best Practices
Track usage with /usage and the status line
Reduce cost: choose the model and effort for the task, manage context and use subagents
Team agreements: which work AI does and which work people must review
Measure AI use in a development team sensibly and keep improving the setup
Section 16: Master Workshop: From Issue to Pull Request
Install hooks and permission rules for the NovaDesk project
Set up GitHub Actions so Claude Code reviews pull requests automatically
Take the "SLA and notifications" issue, then plan, build and test it
Deliver a pull request that passes automated review and is ready to merge
Present the result and wrap up with a checklist for applying the workflow to your own projects
No public rounds are open right now. Join the waiting list and we will contact you first when the next round opens, or ask us on LINE. Or call 02-570-8449 or 088-807-9770
Claude Code Roadmap is a 5-day course that covers Claude Code broadly, AI Coding Agent for SDLC covers several AI coding agents across the development life cycle, and Build AI Multi-Agent with Claude Code focuses on directing advanced agent teams. This course is a 3-day hands-on workshop that delivers one real project from issue to reviewed pull request, with team governance through hooks, permissions, plugins and CI.
The institute does not provide accounts or subscriptions. You use your own or your company's Claude Pro, Max, Team or Enterprise plan, or an API key from the Claude Console. A plan with a smaller allowance works for the course but may reach its usage limit during heavy workshops, while higher-allowance plans are more comfortable for a full day of use. Team or Enterprise seats need Claude Code access from your organisation's admin, and an API key is billed by actual usage, so set a budget. You also need a GitHub account for the pull request and GitHub Actions parts.
No. The main NovaDesk examples use TypeScript, but every concept and technique applies to Python, PHP, Dart, Java or other languages. You can do the workshops in the language you know best, as long as you install that language's runtime and tools yourself.
No. The course starts with installation and setup, but you should be able to program in at least one language, have worked on a multi-file project, use the terminal and Git at a basic level, and understand web APIs and databases, because the course does not teach programming or Git from scratch.
Bring a Windows 10/11, macOS or Linux laptop with at least 8 GB of RAM and permission to install software, with Git, Node.js LTS and VS Code installed in advance. Claude Code runs on Windows, but the sandbox for shell commands works on macOS, Linux and WSL2, so Windows users who want to try that part should set up WSL2 as well.
Not in depth. This course focuses on using Claude Code to develop projects as an individual and as a team, from setup and feature work to extensions, quality control and CI/CD. Building applications with the Claude API or Agent SDK, and writing your own MCP server from scratch, are better suited to specialised follow-on courses.
Articles are published in Thai.
Claude Code is not merely a tool for asking an AI to write code. It acts as an AI coding agent that understands project structure, reads files, edits source code, runs commands, tests the system and works alongside a real development workflow. Anyone still…
Most developers already use AI to help write code, but they use it one task at a time and paste the output back in, which captures value only during the coding step. The reality is that a development team spends more time on planning, breaking down work,…
Many developers already use Claude Code fluently as a coding assistant, one prompt at a time. As the work grows, though, a single agent's context fills up, tasks that could run in parallel are done one after another, and results depend on who writes the best…
