agent-manager_skill

Overview

This skill is an employee agent lifecycle manager that runs AI employee agents inside tmux sessions using only Python and tmux. It provides commands to start, stop, monitor, assign tasks, and schedule agent runs without a central server. The design favors low dependency, cron-friendly scheduling, and clear session isolation for reliable local orchestration.

How this skill works

The tool parses agent definitions from markdown files with YAML frontmatter and maps each agent to a dedicated tmux session named agent-{name}. It launches configured launchers, injects skills as system prompts, and exposes CLI commands for listing, starting, stopping, monitoring, sending messages, assigning tasks, and syncing cron schedules. Scheduling generates crontab entries that trigger non-interactive runs and optionally captures logs.

When to use it

• You need lightweight local orchestration of AI agents without a server or CAO.
• You want each agent isolated in its own tmux session for easy monitoring and interaction.
• You need cron-driven, repeatable jobs for daily tasks, code reviews, or reports.
• You want to assign tasks to Dev/QA agents and observe live output.
• You prefer simple YAML agent configs with skill injection and optional tmux layouts.

Best practices

• Define agents in agents/EMP_*.md with clear name, description, working_directory, and launcher fields.
• Keep tasks concise and prefer task_file for longer scheduled jobs to keep YAML readable.
• Use enabled: false for temporary maintenance or to prevent schedule execution during testing.
• Configure tmux.layout and target_pane when you need deterministic pane placement for multi-pane agents.
• After assign or send, remember to send Enter to the tmux session (tmux send-keys -t agent-{name} Enter) to trigger execution.

Example use cases

• Start a project Dev agent in the morning, assign a ticket, monitor live output, and stop in the evening.
• Schedule nightly QA runs to execute test suites and capture results to log files via cron sync.
• Create a daily-standup job that runs an agent to summarize issues and write a weekly report automatically.
• Assign a one-off bug-fix task to the dev agent from stdin or a task file and follow progress with monitor --follow.
• Run multiple agents (dev, qa) in separate tmux sessions to avoid process pollution and enable direct attachment when needed.