Skip to main content

Claude Code

Angzarr uses Claude Code (Anthropic's CLI tool) for AI-assisted development. Project-specific instructions are maintained in CLAUDE.md files.

CLAUDE.md Files​

Claude Code reads instructions from CLAUDE.md files at multiple levels:

~/CLAUDE.md                    # User-level (global preferences)
~/workspace/angzarr/CLAUDE.md  # Project-level (angzarr-specific)

Project-level instructions override user-level for that codebase.

Angzarr Instructions​

The angzarr CLAUDE.md includes:

Code Patterns​

  • guard/validate/compute — Handler structure for aggregates
  • Error constants — Centralized error messages
  • IoC patterns — Dependency injection via generics

Testing​

  • TDD mandatory — Red/green/refactor cycle
  • Three levels — Unit, integration, acceptance
  • Gherkin specs — Living documentation

Git Practices​

  • No AI attribution — Commits don't mention Claude/Anthropic
  • lefthook — Pre-commit hooks for lint/format/test

SCM Integration​

CLAUDE.md includes assembled context from SCM:

<!-- SCM:BEGIN -->
@.scm/context.md
<!-- SCM:END -->

This keeps AI context in sync with project practices. See SCM for details.

Custom Skills​

Claude Code supports custom skills (slash commands). Angzarr defines:

/nuke-deploy​

Tear down, rebuild from scratch, and redeploy to the Kind cluster.

/code-review​

Comprehensive code review following project standards.

/code-review-recent​

Review recent changes (uncommitted or last commit).

Hooks​

Claude Code hooks execute shell commands in response to events:

{
  "hooks": {
    "pre-commit": ["just check", "just fmt --check"],
    "post-tool-use": {
      "Write": ["just fmt {{file}}"]
    }
  }
}

Hooks ensure code quality without manual intervention.

MCP Servers​

Claude Code can connect to MCP servers for extended capabilities:

{
  "mcpServers": {
    "scm": {
      "command": "scm",
      "args": ["mcp"]
    },
    "mcp-tasks": {
      "command": "mcp-tasks"
    }
  }
}

Available MCPs​

  • scm — Context fragment management
  • mcp-tasks — Task tracking in markdown files

Best Practices​

Keep Instructions Current​

Update CLAUDE.md when project practices change. Stale instructions cause AI confusion.

Be Explicit​

Claude follows instructions literally. Vague guidance produces inconsistent results.

Use Fragments​

Large CLAUDE.md files become unwieldy. Use SCM fragments for modular context.

Test with AI​

After updating instructions, test common tasks to verify AI behavior matches expectations.

Next Steps​