Task Orchestrator

Maintaining complex state across multiple AI sessions is difficult when relying only on context windows. This skill implements a hierarchical task management system via an MCP server, providing persistent, structured storage for goals and subtasks.

What it does

• Goal management creates durable top-level objectives associated with specific repositories.
• Hierarchical tasking uses dot-notation IDs like 1.2.1 to organize work into deeply nested trees.
• Transactional updates ensure task batches are added or modified atomically to prevent state corruption.
• Dependency enforcement prevents marking parent tasks as complete while subtasks remain open.
• Persistent storage utilizes LokiDB to ensure your task lists survive agent restarts or session timeouts.

How it works

1. Initialize a goal using the create_goal tool to establish a new project scope.
2. Build the task tree by passing nested subtask arrays to add_tasks in a single call.
3. Query progress with get_tasks to retrieve first-level or recursive task lists.
4. Update status via complete_task_status, optionally closing entire branches at once.

Frameworks & tools

This skill requires Node.js 18+ and pnpm. It is built as an MCP server using the Model Context Protocol and stores data locally using LokiDB.

Why this beats prompting it yourself

Standard prompts lose task context as conversations grow long, leading to forgotten requirements or duplicate work. This skill moves the "todo list" out of the volatile context window and into a structured database with enforced parent/child logic.

Use cases

• Manage multi-step refactoring projects across several days of work.
• Orchestrate sub-agents by assigning specific dot-notation task IDs.
• Maintain a durable project roadmap that persists through agent crashes.

Known limitations

Does not support goal deletion or inter-goal dependencies. A dedicated progress visualization UI and complex cross-task orchestration are currently not implemented.