mcp server
claude-code-in-action
The process I experienced firsthand through the https://anthropic.skilljar.com/claude-code-in-action course using real-world daily projects
Set up
Runs on your computer. Pick your agent and operating system.
This project does not publish an install method we can turn into a config. Follow the README instead.
README
synced Oct 21, 2026claude-code-in-action
Personal learning repo — real-world Claude Code workflows built and documented through daily use.
Structure
| Folder | Topic | What's inside |
|---|---|---|
00-materials/ | Course materials | Reference PDFs, docs, skill-building guide |
01-small-project/ | Real app (Next.js) | Full-stack app used as the primary sandbox across all lessons |
02-mcp/ | MCP servers | Playwright, repomix, mcporter — browser automation and codebase context |
03-auto-simplify/ | Hook automation | PostToolUse + UserPromptSubmit hooks that enforce /simplify every 5 edits |
04-uiux/ | UI generation | Schemas, hooks, and repomix configs for AI-assisted UI workflows |
05-subagents/ | Subagent patterns | Orchestrator, codebase-analyst, mcp-manager agent setups |
06-github-actions/ | GitHub as event bus | Multi-agent collaboration via GH Actions (@claude, @deepseek triggers) |
07-voice-input/ | Voice → Claude | Whisper-based voice input pipeline to Claude Code |
08-superpowers/ | Superpowers integration | Workflow classifier hook, CC vs Codex analysis, verbosity control |
09-gstack-specific/ | Personal kit | gstack + ecc-tools merged with superpowers; overlap analysis and install guide |
Key Artifacts
03-auto-simplify — hooks that turn /simplify from optional to mandatory. Demonstrates PostToolUse + UserPromptSubmit exit-code mechanics.
06-github-actions — GitHub as a message queue for AI agents. Multiple agents (@claude, @deepseek) on one repo, triggered by issue comments, chaining automatically.
08-superpowers — Why Superpowers doubles Claude Code's capability but reduces Codex's. Includes a workflow-classifier.sh hook that injects context-aware guidelines (debug / plan / review / impl) on every prompt — no changes to CLAUDE.md required.
09-gstack-specific — How to audit, deduplicate, and merge multiple shared skill kits (gstack, ecc-tools, superpowers) into a personal stack without overlap or token bloat.
The Main App (01-small-project)
Next.js + Prisma + Vitest. Used as the real codebase for testing every workflow pattern in this repo.
cd 01-small-project
npm run setup # install, generate Prisma client, run migrations
npm run dev # start on :3000
npm run test # Vitest
Recurring Patterns
- Hooks over config — behavioral enforcement via hooks (
PostToolUse,UserPromptSubmit) is more reliable than longCLAUDE.mdfiles that agents skip - Short injections beat long files — 1-line hook output is read; 200-line config is skimmed
- GitHub as coordination layer — agents don't need direct communication if they share a repo
- Skill kits need auditing — more skills ≠ more capability; overlap causes token bloat and tool confusion