Agent Registry

ClaudeAutoPM uses a core + plugin agent model. 7 core agents are always available. Additional specialized agents (50+) are installed on demand via plugins.

Architecture

The agent registry is defined in .claude/agents/agent-registry.xml. This XML file:

1. Lists all 7 core agents (always present)
2. Contains PLUGIN_AGENTS_START / PLUGIN_AGENTS_END markers
3. During autopm install, the installer injects agents from selected plugins between these markers
4. CLAUDE.md includes this file via @include .claude/agents/agent-registry.xml

This means only agents from your installed plugins consume context tokens.

<agent-registry version="2.0.0" core-agents="7">
  <agents category="core">
    <agent name="agent-manager" path="agents/core/agent-manager.md">...</agent>
    <!-- ... other core agents ... -->
  </agents>

  <!-- PLUGIN_AGENTS_START -->
  <!-- Injected by install.js based on installed plugins -->
  <!-- PLUGIN_AGENTS_END -->
</agent-registry>

Core Agents (Always Available)

These 7 agents are included in every installation scenario, from lite to performance.

agent-manager

Path: .claude/agents/core/agent-manager.mdPurpose: Agent lifecycle management, documentation standards, registry maintenance. Use for: Creating new agents, updating agent definitions, managing the registry.

file-analyzer

Path: .claude/agents/core/file-analyzer.mdPurpose: Analyze and summarize large files to reduce context usage. Use for: Log analysis, documentation review, test output summarization, configuration review.

code-analyzer

Path: .claude/agents/core/code-analyzer.mdPurpose: Code review, bug detection, and logic flow analysis. Use for: Pre-commit reviews, security scanning, architecture analysis, dependency tracing.

test-runner

Path: .claude/agents/core/test-runner.mdPurpose: Test execution and comprehensive result analysis. Use for: Running test suites, failure analysis, coverage reports.

parallel-worker

Path: .claude/agents/core/parallel-worker.mdPurpose: Coordinate multiple agents working on related tasks in a git branch. Use for: Complex multi-step workflows, parallel task execution, multi-file operations.

mcp-manager

Path: .claude/agents/core/mcp-manager.mdPurpose: MCP server install, configuration, process management, health checks. Use for: MCP server setup, configuration, troubleshooting.

context-optimizer

Path: .claude/agents/core/context-optimizer.mdPurpose: Context window efficiency, compaction, summarization, memory optimization. Use for: Reducing context usage, optimizing prompt size, managing long conversations.

Plugin Agents (Installed on Demand)

Plugin agents are only available after installing the corresponding plugin. They are organized across 13 plugins:

Plugin	Agents	Category
plugin-languages	5	Python, Node.js, JavaScript, Bash (see plugin.json)
plugin-frameworks	4	React, Tailwind, UX Design (see plugin.json)
plugin-devops	7	Docker, GitHub Actions, Azure DevOps, Observability, SSH, Traefik (see plugin.json)
plugin-cloud	8	AWS, Azure, GCP, Kubernetes, Terraform, OpenAI, Gemini (see plugin.json)
plugin-databases	6	PostgreSQL, MongoDB, Redis, CosmosDB, BigQuery
plugin-testing	1	Playwright specialist
plugin-ai	8	AI/ML integration agents
plugin-data	3	LangGraph, Airflow, Kedro
plugin-ml	10	Machine learning specialists
plugin-pm	0	PM commands (no agents)
plugin-pm-github	0	GitHub provider (no agents)
plugin-pm-azure	0	Azure provider (no agents)
plugin-core	7	Core agents (always installed)

Installing Plugin Agents

Plugin agents are installed automatically based on your chosen scenario:

# Lite: 7 core agents only
autopm install --scenario=lite

# GitHub: 7 core + 5 language agents
autopm install --scenario=github

# Full: 7 core + 45+ plugin agents
autopm install --scenario=full

Example Plugin Agents

From plugin-languages:

• @python-backend-expert - Python backend development (FastAPI, Django, Flask)
• @nodejs-backend-engineer - Node.js backend with Express, NestJS
• @bash-scripting-expert - Shell scripting and automation

From plugin-devops:

• @docker-containerization-expert - Docker, Compose, multi-stage builds
• @github-operations-specialist - GitHub Actions, CI/CD pipelines

From plugin-cloud:

• @aws-cloud-architect - AWS infrastructure design
• @azure-cloud-architect - Azure resources, ARM templates
• @kubernetes-orchestrator - K8s deployments, Helm charts
• @terraform-infrastructure-expert - Infrastructure as Code

From plugin-databases:

• @postgresql-expert - Schema design, query optimization
• @mongodb-expert - Document design, aggregation pipelines
• @redis-expert - Caching strategies, data structures

Agent Usage

# Core agents work in any scenario
@code-analyzer review this authentication module for security vulnerabilities
@test-runner execute all integration tests and analyze any failures

# Plugin agents require the corresponding plugin
@docker-containerization-expert create optimized Docker setup  # Requires plugin-devops
@postgresql-expert optimize this slow query                     # Requires plugin-databases

Legacy Agent Support

Deprecated agent names were removed in the v1.1.0 consolidation. Use the consolidated versions:

Deprecated	Current
@react-expert	@react-ui-expert (plugin-frameworks)
@docker-expert	@docker-containerization-expert (plugin-devops)
@python-expert	@python-backend-expert (plugin-languages)
@kubernetes-expert	@kubernetes-orchestrator (plugin-cloud)

Related Pages

• Agent Selection Guide - Choose the right agent
• Plugin Development - Create your own plugins
• Architecture - System design overview