Agents

CCW provides specialized agents for different development workflows.

What are Agents?

Agents are specialized AI assistants with specific expertise and tools for different aspects of software development. They are invoked via the Task tool in skills and workflows.

Built-in Agents

Execution Agents

code-developer

Pure code execution agent for implementing programming tasks and writing tests.

Expertise:

• Feature implementation
• Code generation and modification
• Test writing
• Bug fixes
• All programming languages and frameworks

Task({
  subagent_type: "code-developer",
  prompt: "Implement user authentication API",
  run_in_background: false
})

tdd-developer

TDD-aware execution agent supporting Red-Green-Refactor workflow.

Expertise:

• Test-first development
• Red-Green-Refactor cycle
• Test-driven implementation
• Refactoring with test safety

Task({
  subagent_type: "tdd-developer",
  prompt: "Execute TDD task IMPL-1 with test-first development",
  run_in_background: false
})

test-fix-agent

Executes tests, diagnoses failures, and fixes code until tests pass.

Expertise:

• Test execution and analysis
• Failure diagnosis
• Automated fixing
• Iterative test-fix cycles

Task({
  subagent_type: "test-fix-agent",
  prompt: "Run tests and fix any failures",
  run_in_background: false
})

universal-executor

Universal executor for general-purpose execution tasks.

Expertise:

• General task execution
• Document generation
• Multi-step workflows
• Cross-domain tasks

Task({
  subagent_type: "universal-executor",
  prompt: "Generate documentation for the API",
  run_in_background: false
})

Analysis Agents

context-search-agent

Intelligent context collector for development tasks.

Expertise:

• Codebase exploration
• Pattern discovery
• Context gathering
• File relationship analysis

Task({
  subagent_type: "context-search-agent",
  prompt: "Gather context for implementing user authentication",
  run_in_background: false
})

debug-explore-agent

Hypothesis-driven debugging agent with NDJSON logging.

Expertise:

• Root cause analysis
• Hypothesis generation and testing
• Debug logging
• Systematic troubleshooting

Task({
  subagent_type: "debug-explore-agent",
  prompt: "Debug the WebSocket connection timeout issue",
  run_in_background: false
})

cli-explore-agent

CLI-based code exploration agent.

Expertise:

• CLI code analysis
• External tool integration
• Shell-based exploration
• Command-line workflows

Task({
  subagent_type: "cli-explore-agent",
  prompt: "Explore codebase for authentication patterns",
  run_in_background: false
})

Planning Agents

action-planning-agent

Creates implementation plans based on requirements and control flags.

Expertise:

• Task breakdown
• Implementation planning
• Dependency analysis
• Priority sequencing

Task({
  subagent_type: "action-planning-agent",
  prompt: "Create implementation plan for OAuth2 authentication",
  run_in_background: false
})

issue-plan-agent

Planning agent specialized for issue resolution.

Expertise:

• Issue analysis
• Solution planning
• Task generation
• Impact assessment

Task({
  subagent_type: "issue-plan-agent",
  prompt: "Plan solution for GitHub issue #123",
  run_in_background: false
})

Specialized Agents

team-worker

Unified team worker agent for role-based collaboration.

Expertise:

• Multi-role execution (analyst, writer, planner, executor, tester, reviewer)
• Team coordination
• Lifecycle management
• Inter-role communication

Task({
  subagent_type: "team-worker",
  description: "Spawn executor worker",
  team_name: "my-team",
  name: "executor",
  run_in_background: true,
  prompt: "## Role Assignment\nrole: executor\n..."
})

doc-generator

Documentation generation agent.

Expertise:

• API documentation
• User guides
• Technical writing
• Diagram generation

Task({
  subagent_type: "doc-generator",
  prompt: "Generate documentation for the REST API",
  run_in_background: false
})

Agent Categories

Category	Agents	Purpose
Execution	code-developer, tdd-developer, test-fix-agent, universal-executor	Implement code and run tasks
Analysis	context-search-agent, debug-explore-agent, cli-explore-agent	Explore and analyze code
Planning	action-planning-agent, issue-plan-agent, cli-planning-agent	Create plans and strategies
Specialized	team-worker, doc-generator, ui-design-agent	Domain-specific tasks

Agent Communication

Agents can communicate and coordinate with each other:

// Agent sends message
SendMessage({
  type: "message",
  recipient: "tester",
  content: "Feature implementation complete, ready for testing"
})

// Agent receives message via system

Team Workflows

Multiple agents can work together on complex tasks:

[analyst] -> RESEARCH (requirements analysis)
    |
    v
[writer] -> DRAFT (specification creation)
    |
    v
[planner] -> PLAN (implementation planning)
    |
    +--[executor] -> IMPL (code implementation)
    |               |
    |               v
    +-----------[tester] -> TEST (testing)
    |
    v
[reviewer] -> REVIEW (code review)

Using Agents

Via Task Tool

// Foreground execution
Task({
  subagent_type: "code-developer",
  prompt: "Implement user dashboard",
  run_in_background: false
})

// Background execution
Task({
  subagent_type: "code-developer",
  prompt: "Implement user dashboard",
  run_in_background: true
})

Via Team Skills

Team skills automatically coordinate multiple agents:

Skill({
  skill: "team-lifecycle",
  args: "Build user authentication system"
})

Configuration

Agent behavior is configured via role-spec files in team workflows:

---
role: executor
prefix: IMPL
inner_loop: true
subagents: [explore]
---

See Also

• Skills - Reusable skill library
• Workflows - Orchestration system
• Teams - Team workflow reference