Skip to content
⍼ Angzarr
Angzarr Blog

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

Section titled “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

Section titled “Angzarr Instructions”

The angzarr CLAUDE.md includes:

Code Patterns

Section titled “Code Patterns”
  • @handles / @applies — Decorator-driven OO aggregate structure
  • Error constants — Centralized error messages
  • IoC patterns — Dependency injection via generics

Testing

Section titled “Testing”
  • TDD mandatory — Red/green/refactor cycle
  • Three levels — Unit, integration, acceptance
  • Gherkin specs — Living documentation

Git Practices

Section titled “Git Practices”
  • No AI attribution — Commits don’t mention Claude/Anthropic
  • lefthook — Pre-commit hooks for lint/format/test

SCM Integration

Section titled “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

Section titled “Custom Skills”

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

/nuke-deploy

Section titled “/nuke-deploy”

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

/code-review

Section titled “/code-review”

Comprehensive code review following project standards.

/code-review-recent

Section titled “/code-review-recent”

Review recent changes (uncommitted or last commit).

Hooks

Section titled “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

Section titled “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

Section titled “Available MCPs”
  • scm — Context fragment management
  • mcp-tasks — Task tracking in markdown files

Best Practices

Section titled “Best Practices”

Keep Instructions Current

Section titled “Keep Instructions Current”

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

Be Explicit

Section titled “Be Explicit”

Claude follows instructions literally. Vague guidance produces inconsistent results.

Use Fragments

Section titled “Use Fragments”

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

Test with AI

Section titled “Test with AI”

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

Next Steps

Section titled “Next Steps”