Skip to main content

Domain 3 — Claude Code Configuration & Workflows

Exam weight: 20%

This domain tests your ability to configure Claude Code for team workflows using CLAUDE.md files, Agent Skills, custom slash commands, hooks, and CI/CD integration.

What this domain tests​

Task StatementDescription
3.1Configure persistent project context with CLAUDE.md
3.2Design and use plan mode for high-impact operations
3.3Create custom slash commands for repeatable workflows
3.4Integrate Claude Code into CI/CD pipelines
3.5Apply hooks for deterministic post-action behavior
3.6Configure tool access with allowedTools and permissionMode

CLAUDE.md — the persistent project context​

CLAUDE.md is a markdown file that Claude Code reads automatically at the start of every session. It functions as a "tech lead in a file" — encoding institutional knowledge that applies to every interaction.

Hierarchy (most specific wins)​

subdirectory CLAUDE.md  ← highest priority
       ↑
project root CLAUDE.md
       ↑
~/.claude/CLAUDE.md     ← lowest priority (global user defaults)

Example: monorepo structure

/project/
  CLAUDE.md              ← shared rules: git conventions, PR format, CI commands
  /frontend/
    CLAUDE.md            ← React patterns, component conventions, design system tokens
  /backend/
    CLAUDE.md            ← Python type hints, FastAPI patterns, test structure

What belongs in CLAUDE.md​

# Project: Payments Platform

## Build commands
- `npm run dev` — start dev server
- `npm test` — run full test suite
- `npm run lint && npm run typecheck` — required before PR

## Architecture
- Services communicate via gRPC, not REST
- All new services must implement health check at /health
- Database: Postgres 15 via Prisma ORM

## Coding standards
- TypeScript strict mode required everywhere
- No `any` types — use `unknown` and type guards
- All async functions must handle errors explicitly

## PR checklist
- [ ] Tests added for new behavior
- [ ] No console.log left in code
- [ ] API changes documented in /docs/api

Plan mode​

Plan mode shows Claude's full intended action sequence before any file is modified. It is the human review checkpoint for high-stakes changes.

When plan mode is required:

ScenarioWhy
Large-scale refactor (10+ files)Verify scope before execution
Database schema migrationIrreversible in production
CI/CD deploymentBroad system impact
Authentication changesHigh-risk, hard to reverse
# Invoke plan mode
/plan

# Or with a specific task description
/plan refactor authentication to use JWT across all services

Claude shows the plan → you review → approve, modify, or cancel.

Custom slash commands​

Custom slash commands package repeatable multi-step workflows:

/review-pr     → lint + test coverage + security scan + summarize findings
/add-tests     → explore structure + identify gaps + generate tests + run them
/deploy-staging → build + test + deploy + smoke test + notify team

Commands are defined in .claude/commands/ and can be shared across the team via version control.

Hooks​

Hooks run shell commands at specific points in the Claude Code agent loop:

// .claude/settings.json
{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": { "tool_name": "Write" },
        "hooks": [{ "type": "command", "command": "npm run lint -- ${file}" }]
      }
    ]
  }
}

This hook runs the linter automatically after every file write — deterministic quality enforcement without prompting.

Tool access configuration​

# Read-only exploration agent
options = ClaudeAgentOptions(
    allowed_tools=["Read", "Glob", "Grep"],
    permission_mode="default"  # prompts before any write
)

# Development agent with auto-approved edits
options = ClaudeAgentOptions(
    allowed_tools=["Read", "Glob", "Grep", "Write", "Bash"],
    permission_mode="acceptEdits"  # auto-approves file operations
)

# Explicitly block sensitive tools
options = ClaudeAgentOptions(
    allowed_tools=["Read", "Glob", "Grep", "Write"],
    disallowed_tools=["Bash"]  # no shell execution
)

permissionMode values:

  • "default" — prompts user for approval on each write operation
  • "acceptEdits" — auto-approves all file read/write; appropriate for CI/CD

CI/CD integration​

Claude Code in CI pipelines runs automated review on every PR:

# .github/workflows/claude-review.yml
- name: Run Claude Code review
  run: |
    claude -p "Review this PR for: security issues, missing tests, \
    architecture violations per CLAUDE.md, and documentation gaps. \
    Output structured JSON with findings array." \
    --output-format json > review.json

Key considerations for CI prompts:

  • Load CLAUDE.md automatically (happens by default)
  • Request structured output for downstream processing
  • Include specific review criteria, not "review this code"
  • Calibrate false positive rate with project-specific examples in CLAUDE.md

Official documentation​