long task manager

The problem

LLM session resets, crashes, and context compression wipe your agent's memory. When long-running tasks like data migrations or web crawls span multiple days, the agent loses track of progress, background process IDs, and intermediate artifacts.

What it does

• Generates persistent JSON state files in a dedicated task directory to serve as a durable source of truth.
• Manages background process lifecycles by tracking session IDs, monitoring logs, and handling process recovery.
• Implements a strict resume protocol that allows fresh agent sessions to pick up exactly where a previous one left off.
• Provides a Python-based helper script for atomic updates, schema validation, and checkpointing.

Frameworks & tools

Compatible with any agent host providing terminal access. Utilizes Python 3 for state management and standard Unix/Windows process controls.

Why this beats prompting it yourself

Standard prompting relies on context window memory which is volatile. This skill enforces a filesystem-first discipline, ensuring that even if an agent is restarted or the context is flushed, the "next action" and background process pointers remain executable and valid.

Use cases

• Executing multi-hour database migrations or index rebuilds.
• Managing background web scraping or data processing jobs that outlive the session.
• Coordinating multi-step research tasks that require human feedback over several days.
• Recovering task context after a system crash or gateway restart.

Known limitations

Background processes started via terminal calls do not survive host restarts. For persistence across reboots, systemd or cron integration is required.