Lessons in Claude Code

The opinionated short-list: auto-run dotnet format after every file edit, block dangerous EF Core commands like dotnet ef database update before they execute, and force tests after every implementation. Why CLAUDE.md instructions are advisory but hooks are deterministic - and which guardrails earn a place in a real .NET repo.

Understand the Model Context Protocol and how MCP servers extend Claude Code's capabilities. Connect to databases, APIs, and third-party services.

Create your own MCP server in C# with the official SDK - a complete, runnable .NET 10 server with DI, logging, auth, and a real tool that exposes your Web API's capabilities to Claude. Not a weather demo.

Wire up Postgres and SQL Server MCP servers so Claude sees your real schema while writing EF Core queries and migrations - and learn when schema-aware AI helps vs when it confidently writes the wrong LINQ.

Connect Claude to your logs and traces - Seq, Application Insights, Grafana - so it triages incidents with real telemetry instead of pasted snippets. Covers the MCP setup and the triage prompts that actually surface root causes.

Leverage sub-agents to parallelize tasks and delegate complex operations. Understand when and how to use multi-agent workflows in .NET projects.

Run Claude Code without the interactive terminal: scripted runs with claude -p, structured JSON output, exit codes, cost controls, and automated triage of failing .NET builds.

Wire Claude Code into GitHub Actions to review every pull request. Includes the workflow YAML, a review prompt tuned for .NET smells like async-over-sync and EF Core N+1, and an honest 30-day scorecard of what it caught vs missed.

Set up the workflow where Claude picks up a labeled GitHub issue and opens a reviewed draft PR: the automation setup, the guardrails that keep it safe, and which issue types it handles well vs where a human should start.

Scheduled headless Claude runs that keep a .NET codebase healthy while you sleep: NuGet dependency bumps, new analyzer warning fixes, and documentation drift checks - each delivered as a small, verified PR.

The senior deliverable is shifting from code to specs. Learn the ADR and spec template that produces correct .NET implementations - demonstrated by running the same feature from a weak spec vs a strong one.

Defense in depth for AI-assisted codebases: CLAUDE.md conventions as suggestions, Claude Code hooks as gates, and Roslyn analyzers plus architecture tests as law. Watch Claude attempt a violation and get caught at each layer.

Turn the .claude folder into a team asset: shared settings and permission allowlists, skills as team conventions, repo vs user scope, and onboarding a new developer with a single clone.

Package your team's skills, hooks, agents, and MCP configuration into a private plugin so every repository and every developer gets the same setup - versioned, distributable, and maintained in one place.

The judgment capstone: race conditions and async coordination, transaction scope across services, EF migrations on live databases, time zone math, and distributed consistency - with a real failing example for each.