← Back to Context Comparison

User-Level Agents

Specialized autonomous agents loaded from the user's home directory

Location: ~/.claude/agents/
1 Agent Universal Scope Multi-Agent Orchestration

What Are User-Level Agents?

User-level agents are autonomous specialists stored at ~/.claude/agents/. Unlike skills (which provide knowledge), agents are invoked via the Task tool to perform complex, multi-step operations autonomously. They can coordinate with other agents, use tools, and produce comprehensive deliverables. The PM agent at this level provides universal project orchestration capabilities across all projects.

📋

Project Manager (PM) Sonnet

pm

Project orchestration specialist for multi-step tasks requiring coordination between specialized agents. Breaks down complex projects into manageable tasks, delegates to architect/security/developer agents, tracks progress through file-based communication, and synthesizes results into coherent deliverables.

🛠️ Available Tools

Read Write Grep Glob Bash

🎯 Core Expertise

  • Task decomposition and delegation to specialist agents
  • Multi-agent workflow coordination
  • Progress tracking and status reporting
  • Risk identification and mitigation
  • Resource allocation and timeline management
  • Cross-functional team communication
  • Stakeholder reporting and documentation

🤖 Coordinates With

  • Architect Agent - System design, technology stack, integration planning
  • Security Agent - Network topology, firewall, SSL/TLS, OAuth2/Keycloak
  • Developer Agent - Implementation, deployment, debugging, containers

📂 Workspace Structure

  • ~/projects/data/claudeagents/pm/tasks/ - Task breakdowns and assignments
  • ~/projects/data/claudeagents/pm/reports/ - Progress summaries
  • ~/projects/data/claudeagents/pm/notes/ - Coordination notes
  • ~/projects/data/claudeagents/pm/escalations/ - Inter-agent escalations

📝 Typical Workflow

1 Analyze user request and break into logical subtasks
2 Write task breakdown to tasks/{project-path}/
3 Invoke Architect agent for design decisions
4 Invoke Security agent for network/auth configuration
5 Invoke Developer agent for implementation
6 Monitor progress via agent output files
7 Synthesize final report to reports/{project-path}/

📚 Context Files

  • ~/.claude/agent-docs/pm/infrastructure-overview.md - System state
  • ~/.claude/agent-docs/pm/project-templates.md - Task templates
  • ~/.claude/agent-docs/pm/agent-coordination.md - Coordination patterns
  • ~/.claude/agent-docs/INTER-AGENT-PROTOCOL.md - Communication protocol

User-Level vs Project-Level Agents

Aspect User-Level Agents Project-Level Agents
Location ~/.claude/agents/ ~/projects/.claude/agents/
Scope Universal orchestration Infrastructure-specific implementation
Count 1 agent (PM) 4 agents (Architect, Developer, Security, QA)
Role Coordinates other agents Execute specialized tasks
Invocation Task tool with subagent_type="pm" Task tool with subagent_type="architect|developer|security"

Related: User-Level Skills | Project-Level Skills | Context Architecture

Agents are auto-discovered by Claude Code from ~/.claude/agents/