agent-manager

Agent lifecycle management via tmux + Python.

Repo: fractalmind-ai/agent-manager-skillLanguage: Python Status: Stable Install: npx openskills install fractalmind-ai/agent-manager-skill

What It Does

agent-manager is the L0 management layer — it handles everything about individual AI agents:

• Start/Stop: Launch agents in tmux sessions
• Monitor: Check agent health and status
• Heartbeat: Periodic health checks with configurable frequency
• Schedule: Cron-based task scheduling
• Assign: Give tasks to specific agents

Key Design

• Zero server: Uses tmux sessions, no HTTP API needed
• File-based config: Agent definitions in YAML files
• Cron-friendly: Heartbeats and schedules via system crontab
• Observable: Attach to any agent session in real-time

Usage

Load the skill in your AI agent:

npx openskills read agent-manager

Then your agent can manage other agents:

# Start an agent
python3 scripts/main.py agent start EMP_0003

# Check status
python3 scripts/main.py agent list

# Assign a task
python3 scripts/main.py agent assign EMP_0003 --task "Review PR #123"

# Sync heartbeat to crontab
python3 scripts/main.py heartbeat sync

Agent Configuration

Agents are defined in YAML frontmatter in agents/*.md files:

---
name: dev-agent
description: Development agent for code tasks
working_directory: /path/to/workspace
heartbeat:
  cron: "*/30 * * * *"
  enabled: true
skills:
  - agent-manager
  - okr-manager
---

Where It Fits

agent-manager is the foundation of the management layer:

fractalbot → agent-manager (L0) → team-manager (L1)
                   │
                   ├── heartbeat monitoring
                   ├── task assignment
                   └── agent lifecycle

All other management skills depend on or integrate with agent-manager.

Related Components

• team-manager — builds on agent-manager for team coordination
• okr-manager — integrated via heartbeat
• fractalbot — routes messages to agents