task-engine_skill

Overview

This skill is a multi-agent task orchestration engine that manages complex, multi-step projects with a state machine and JSON-based persistence. It dispatches subtasks to different agents, tracks progress via heartbeats, and formats progress and alerts for Discord. The engine includes CLI utilities for creating, transitioning, dispatching, checking, and archiving tasks. It focuses on reliability, clear state transitions, and lightweight heartbeat checks for automated monitoring.

How this skill works

Tasks are stored as JSON directories and indexed by an index.json file. Each task contains subtasks that the engine dispatches to agents (e.g., claude-code, eva) according to agent capabilities, dependencies, and priority. A heartbeat check scans active tasks, applies auto-transitions, detects stuck/failed subtasks, and emits alerts; notifications are rendered as plain-text Discord messages. CLI commands provide machine-readable JSON output for scripting and automation.

When to use it

• Coordinating complex projects that require multiple specialized agents or tools.
• Automating monitoring and progress tracking for multi-step workflows.
• Enforcing deterministic state transitions and history for multi-agent work.
• Generating Discord-friendly progress digests and urgent alerts.
• Recovering or auditing task state with rebuild and archive features.

Best practices

• Model work as tasks with clear subtasks and explicit dependencies to enable auto-dispatch.
• Tag subtasks with types (dev, test) so auto-transitions move tasks through TESTING and REVIEW correctly.
• Use --json output for integrations and cron jobs to parse machine-readable results.
• Integrate the heartbeat check into your system cron/beat to catch stuck subtasks and overdue tasks early.
• Limit parallel agent use according to agent capacities to avoid dispatch failures.

Example use cases

• Implementing a feature rollout: split work into dev and test subtasks, dispatch to claude-code and eva, and monitor transitions to REVIEW and COMPLETED.
• Continuous integration orchestration: auto-dispatch test subtasks and send Discord alerts when a subtask fails or is stuck.
• Managed migration projects: track multi-phase plans with ETA alerts and archive completed phases for auditing.
• On-call incident playbooks: assign diagnostic and remediation subtasks to agents and notify via Discord digest.