Skip to main content

Claude Code and slash commands: agentic development from A to Z

Move from occasional Claude Code usage to an engineering practice: know every slash command and when to type it, teach the project to Claude once and for all, write your own commands, lock down what needs locking with hooks and permissions, parallelize with subagents, wire the team's tools through MCP, package the whole thing as a plugin, and take it all the way to CI.

Course duration: 8h​

What you will learn​

  • Install Claude Code on every surface (terminal, VS Code, JetBrains, desktop app, web) and read what a session actually costs.
  • Understand the agentic loop, the built-in tools and the context window, then drive them with /context, /compact and /clear.
  • Write a CLAUDE.md and targeted rules that hold up over time, and know what the automatic memory retains.
  • Know every built-in slash command, grouped by family, with each alias and each argument.
  • Build your own commands as skills: full frontmatter, arguments, shell output injection, execution inside a subagent.
  • Configure permissions, execution modes, sandbox and settings files without blocking the team or opening a hole.
  • Work without fear thanks to plan mode, checkpoints and worktrees.
  • Automate and lock Claude's behavior with deterministic hooks.
  • Delegate to subagents, launch background work, and orchestrate large-scale changes.
  • Connect Claude Code to GitHub, a database and internal tools through MCP, then ship the whole thing as a plugin.
  • Run Claude Code without a UI: claude -p, GitHub Actions, scheduled routines and the Agent SDK.
  • Master costs, prevent prompt injection, and deploy Claude Code inside an organization.

Prerequisites​

  • Comfortable in a terminal and with Git (branches, commits, pull requests).
  • Able to read Python and TypeScript; the running example uses FastAPI and React, but no framework expertise is required.
  • A Claude account (Pro, Max, Team or Enterprise subscription) or an Anthropic API key to run the exercises.
  • Course 17 (prompt engineering) and course 30 (AI agents) clarify the "why" behind several mechanisms, but are not required.

Course modules​

  1. Install Claude Code and nail your first session
  2. The agentic loop, built-in tools and the context window
  3. CLAUDE.md, rules and memory: teach the project to Claude
  4. Built-in slash commands (1/2): session, context, configuration and model
  5. Built-in slash commands (2/2): quality, parallelism, automation and integrations
  6. Build your own slash commands: skills from A to Z
  7. Permissions, execution modes, sandbox and settings files
  8. Plan mode, checkpoints, sessions and worktrees: working without fear
  9. Hooks: automate and lock Claude's behavior
  10. Subagents, parallel agents and agent teams
  11. MCP: connect Claude Code to your tools and your data
  12. Plugins and marketplaces: package and share your tooling
  13. Daily workflows: explore, fix, test, review, ship
  14. Non-interactive mode, GitHub Actions CI and the Agent SDK
  15. Cost, security and team deployment at scale
  16. Project: the complete Claude Code toolkit for the Kiosque team

Running example​

Kiosque, an order-taking app for food trucks inherited from a contractor, runs through all sixteen modules: a FastAPI API, a small React front-end, tests at 41% coverage, no written conventions, and a .env still lingering in the repo. You are in charge of tooling the team — Nadia, Karim and Lea — with Claude Code. Module after module, Kiosque gains a CLAUDE.md and rules, six in-house slash commands, permissions and a sandbox, three hooks, two subagents, two MCP servers, a kiosque-tools plugin, a CI that answers @claude, and module 16 assembles everything into a toolkit that you can transpose to your own team the following week.

Why this course is different​

The content follows the official Claude Code documentation as of September 2026: custom commands turned into skills, automatic permission mode, forks and subtasks, dynamic workflows, the Agent SDK. Every command, every frontmatter key and every hook event we mention actually exists; whenever a behavior depends on the version, the course says so. You are not learning "how to chat with an AI", you are learning an engineering tool, with its configuration files, its safety rails and its costs.

Assessment and certificate​

The course ends with a 40-question exam covering all sixteen modules. On success, a certificate of completion is issued immediately; its number is verifiable on the platform.

Free courses, by contrast, end with a 5-question quiz and a preview of the certificate, without certification.