task-orchestrator_skill

Overview

This skill is an autonomous task orchestrator that drives work from todo to completed using strict quality gates. It manages state transitions, subtasks, validation loops, and review handoffs to ensure reliable deliveries. The agent enforces checks like builds and tests before marking tasks done. It is built for TypeScript workflows and integrates with tooling for execution and validation.

How this skill works

The orchestrator claims tasks, locks context, and updates status to reflect reality before work begins. It executes or splits tasks, runs validation (type checks, build, tests), and triggers code review when validation passes. If reviews fail or validation fails, it creates fix subtasks and re-queues work until all checks and subtasks pass. Completion only happens after implementation, validation, review approval, and subtask closure.

When to use it

• Automating developer task flow from assignment to completion.
• Enforcing CI-like validation and review gates as part of task lifecycle.
• Coordinating complex tasks that require splitting into subtasks.
• Preventing premature task completion without build/test confirmation.
• Managing status truth and context-sensitive execution based on PRD requirements.

Best practices

• Always read linked PRD or requirement fields before starting a task.
• Mark status to in-progress when work begins and update status immediately on state changes.
• Run full project validation (types, build, tests) before requesting review.
• Split complex work into smaller subtasks to avoid blocking the main task.
• Create explicit fix subtasks when reviews fail instead of modifying the main task silently.

Example use cases

• Claiming the next todo task, loading the PRD, and locking the item for work.
• Splitting a large feature into multiple subtasks and orchestrating their execution and validation.
• Running a validation loop: type check, build, and tests, then triggering code-review on success.
• Automatically creating small fix subtasks after a failed code review and re-queuing them.
• Using helper scripts to inspect a task tree and confirm all subtasks are completed before finalizing.