Claude Code Agents

This project includes six specialized subagents located in .claude/agents/. Each agent is invoked via the Agent tool or by Claude Code automatically when the task matches the agent's description.

Available Agents

How to Use

Invoking an agent in a prompt

Ask Claude Code to use a specific agent by name:

Use the test-writer agent to write tests for src/controllers/authController.js

Run the code-reviewer agent on the latest changes in src/routes/

Have the compliance-reviewer audit src/middleware/ for GDPR issues

Automatic invocation

Claude Code will automatically select the right agent when the task description matches. For example, asking "write tests for this file" may automatically route to test-writer.

Chaining agents (via agent-watcher)

The agent-watcher agent is designed to orchestrate multi-agent workflows:

Use agent-watcher to coordinate a full review: run code-reviewer and 
compliance-reviewer on src/auth/, then synthesize their findings.

Agent Details

test-writer

• Model: claude-sonnet-4-6
• When to use: Adding test coverage, writing tests for new features, generating test suites for existing untested code
• Input: A file path, function, or module to test
• Output: A complete, runnable test file

code-reviewer

• Model: claude-sonnet-4-6
• When to use: Pre-merge reviews, auditing a PR diff, checking a new file before committing
• Input: A file, diff, or set of files to review
• Output: Structured review with Critical / Important / Minor findings

doc-generator

• Model: claude-haiku-4-5-20251001
• When to use: Generating README files, adding JSDoc/TSDoc comments, creating API reference docs
• Input: A file, module, or API surface to document
• Output: Ready-to-use documentation in the project's existing format

architecture-reviewer

• Model: claude-sonnet-4-6
• When to use: Evaluating a new system design, reviewing module boundaries, assessing a significant refactor
• Input: A description of the system, or a set of files representing an architectural layer
• Output: Structured assessment with strengths, concerns (High/Medium/Low), and recommendations

compliance-reviewer

• Model: claude-sonnet-4-6
• When to use: Security audits, GDPR/privacy reviews, checking for OWASP vulnerabilities, license scanning
• Input: Files, configs, or directories to audit
• Output: Findings grouped by compliance area with severity, location, regulation, and remediation

agent-watcher

• Model: claude-opus-4-6
• When to use: Multi-agent workflows, quality-gating other agents' outputs, synthesizing findings from several reviews
• Input: Outputs from one or more other agents, or a complex task requiring coordination
• Output: Cross-agent assessment with conflicts, gaps, and a final verdict on whether work is ready to act on

File Structure

.claude/
└── agents/
    ├── AGENTS.md               ← this file
    ├── test-writer.md
    ├── code-reviewer.md
    ├── doc-generator.md
    ├── architecture-reviewer.md
    ├── compliance-reviewer.md
    └── agent-watcher.md